We need to translate Russian news post into English, format per rules.

First line: short headline under 90 chars, no markdown.

Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.

Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post.

Wrap study/journal citations and publication references in single underscores _ (e.g., Nature Aging, July 2026).

We need to preserve facts, numbers, names, citations exactly.

Let's extract facts:

- scPILOT transfers response of cells to a known stimulus to another patient, cell type, or species.
- Article published August 29 in Advanced Science about scPILOT, a computational model that uses measured cellular response to a stimulus to predict response in another patient, cell type, or species.
- In tests, authors hid cells after stimulus exactly where they made predictions: models only had baseline cells, hidden measurements served as final test.
- Single-cell sequencing shows which genes are active in individual cells.
- Experiment measures one group of cells before stimulus, another after.
- Therefore ordinary calculation often reduces response of whole group to average difference and loses cell-to-cell differences.
- Authors apply optimal transport — a mathematical way to match probabilistic pairs between two groups with minimal total discrepancy. It links cells before and after already measured stimulus and estimates for each baseline cell the shift in gene activity.
- Then model receives baseline cells of a new patient, another cell type, or another species. It finds in prior experiments groups with similar gene activity, transfers their cellular shifts to the new group, and weights more heavily the closer groups.
- Adding this shift to baseline state, scPILOT predicts gene activity after same stimulus.
- On seven immune cell types treated with interferon-beta, a signaling protein of the immune system, scPILOT obtained average R² 0.945.
- This metric shows how predicted average gene activity levels matched measured; best of remaining models had 0.935.
- In test on eight patients R² was 0.598, roughly like CellOT, and by distributions of individual cells scPILOT showed smallest discrepancy.
- Accuracy depended on proximity of baseline cell states.
- Three most similar contexts gave almost same accuracy as all available contexts together, while three most distant worsened prediction.
- On six cancer cell lines differences between lines were stronger than effect of gene knockout. There scPILOT by average gene activity levels was comparable to simple baseline prediction.
- scPILOT uses past experiments as source of prediction, matching their cells to new population: the closer the baseline states, the more useful the transferred response.

We need to include citations: The article published Aug 29 in Advanced Science. So citation: Advanced Science, August 29 maybe? Actually they said "29 августа в Advanced Science вышла статья о scPILOT". So we can wrap as Advanced Science, August 29.

Also maybe mention interferon-beta as a drug/protein: interferon-beta.

Numbers: R² 0.945, 0.935, 0.598, seven immune cell types, eight patients, six cancer cell lines, three most similar contexts, three most distant.

We need at most 4-5 double asterisk facts. Choose key numbers: R² 0.945, R² 0.935, R² 0.598, seven immune cell types, eight patients, six cancer cell lines, interferon-beta maybe. That's many. Need to limit to 4-5. Let's pick: 0.945, 0.935, 0.598, seven immune cell types, eight patients. Could also include interferon-beta as a protein name but maybe not needed. We'll keep 5.

Wrap each in double asterisks.

Now paragraphs: short paragraphs 2-3 sentences each.

Let's craft:…

🔗 Read original →

Beijing, Tianjin and Hebei Release Joint Neurointerface Standards Plan

On August 26, regulators from Beijing, Tianjin, and Hebei published a joint plan for common standards and testing of neurointerfaces. It outlines rules for neural signals and devices, measuring accuracy and safety, and testing sites.

A neurointerface connects the brain to a device; electrodes record brain electrical activity and send the signal to software, which recognizes a command and makes the device act or return feedback. To compare systems, one must check signal recording, algorithm performance, device properties, and the whole chain.

In the full plan, the three regions assign future work along this chain: electrodes, chips for recording and decoding neural signals, recording systems, and real‑time algorithms. The document tasks them with developing data‑quality and component requirements, device characteristics, safety, and ethics.

One task is to create rules for labeling EEG data—records of brain electrical activity. The plan also calls for assessing signal‑collection accuracy and long‑term safety of neurointerfaces. Common rules should define how to describe these recordings, and how to measure and test components and whole systems.

The plan includes work on quality‑standard laboratories, a measurement and certification center, testing tools, and safety, efficacy, and reliability metrics. Test sites should gather EEG databases, conduct clinical studies, and verify technologies in applied scenarios.

These tasks form a sequence: record the signal, measure system properties, and check application. A 2025 joint document of seven Chinese ministries set a 2027 goal to develop neurointerface technology, industry, and standards; the regional plan translates that goal into work on signals and data, device measurement, testing, and research sites.

🔗 Read original →

Google unveils PPE for automated geographic forecasting

Google unveiled its Planetary Prediction Engine (PPE) on August 27. The system receives a textual task and labeled source data, then produces a geographic forecast. It is designed to automate the data‑gathering and preparation steps that epidemiologists normally perform manually.

According to the technical preprint, preparing data for an operational epidemic forecast involves more than >700 actions. PPE first defines the target area and time window and decides how to link tables. It then searches open geographic databases, government portals and scientific repositories, harmonizes the information to common district boundaries and assembles a training set.

Using that training set, PPE builds a forecast and tests it on held‑out districts and weeks. During the Bundibugyo ebolavirus outbreak in the Democratic Republic of Congo, the system ranked districts each week before any cases were reported. Over five weekly forecasts it placed 15 of 18 districts that later saw their first cases inside its top‑ten risk list, achieving 83.3% accuracy.

The authors compare this result with the approximately ~73% performance of an earlier published model that also ranked ten districts by risk. PPE separates training from verification; its Feature Gate filter removes variables that would pre‑reveal the answer, such as part of the target indicator, same‑questionnaire data, event consequences or future information. Missing values are imputed only on the training

🔗 Read original →

We need to translate Russian news post into natural English, format according to rules.

First line: short headline under 90 chars, no markdown, no '#'.

Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.

Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post,5 per post, never a whole sentence.

Wrap study/journal citations and publication references in single underscores _ (e.g., Nature Aging, July 2026).

We need to preserve all facts, numbers, names and citations exactly. No added commentary.

We need to identify important facts: dates: 29 August (the workshop launched pilot). Also maybe "AIDaRS" name, "OpenReview", "GitHub", "NeurIPS". Also maybe "AIDaR workshop at NeurIPS". Also maybe "technical report AIDaRS". Also maybe "voluntary pilot". Also maybe "closed GitHub repositories". Also maybe "pull request". Also maybe "reviewer can leave general feedback or comment on a line". Also maybe "author receives via service". Also maybe "pilot will test whether GitHub review tools give more useful and reproducible feedback". Also maybe "organizers want to know whether uniformly structured project is easier for humans and software agents". Also maybe "participation voluntary, OpenReview remains mandatory submission route".

We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. So we can wrap 29 August maybe. Also maybe NeurIPS? That's a conference name, not a number but maybe considered important fact. The rule says "key numbers, percentages, drug/company/gene names, dates". So conference name not required but could be considered important fact? It says "key numbers, percentages, drug/company/gene names, dates". So we should limit to those. So we can wrap date 29 August. Also maybe OpenReview? It's a service name, not a number/percentage/drug/company/gene name/date. So maybe not. However we could wrap GitHub? It's a company name, qualifies as company name. So we can wrap GitHub in double asterisks. Also maybe AIDaRS? It's a project name, not a company/drug/gene. Not required. Could wrap NeurIPS? It's a conference, not a company/drug/gene. Not required. But we could still wrap if we want but limit to 4-5. Let's pick: 29 August, GitHub, OpenReview (maybe not needed), NeurIPS (maybe not needed). Let's stick to date and GitHub as two. Could also include "AIDaRS" as a project name but not required. Might be okay to wrap it but risk exceeding limit. Let's keep to 2-3.

We also need to wrap study/journal citations and publication references in single underscores _. There's no explicit citation like a journal article; but there is "technical report AIDaRS". That's not a journal citation. There's no explicit study/journal citation. So we may not need any underscores. But we could wrap "technical report AIDaRS" maybe not needed. The rule says wrap study/journal citations and publication references in single underscores. If none, we don't add.

Thus we will have double asterisks around 29 August and GitHub maybe also OpenReview? It's a service name, not a company? It's a company/platform. Could be considered company name. But to be safe, we can wrap GitHub and OpenReview as company names. That's two. Plus date. That's three.

Now format: Headline line, blank line, then paragraphs.

We need to split into short paragraphs 2-3 sentences each.…

🔗 Read original →

AdaptiveFlow screens 69 billion compounds to find new FSP1 and PARP1 inhibitors

On September 1, the journal Nature Biotechnology published an article describing the open‑source virtual‑screening system

🔗 Read original →

We need to translate Russian news post into natural English, format per rules.

First line: short headline under 90 chars, no markdown, no '#'.

Then blank line, then body paragraphs (2-3 sentences each), separated by blank lines.

Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post.

Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026).

We need to preserve facts, numbers, names, citations exactly. No added commentary.

We need to identify important facts: numbers like 29 August, seven methods, two datasets, 27 combinations, 18 of 27, 12, etc. Also Monroe + TabPFN, CheMeleon, Mol-JEPA + TabICL, MEGA-CL, ExpansionRx, Biogen ADME, 18 combinations, etc. Also Claude Code, 22 human queries.

We need to be selective: at most 4-5 double asterisks. So we need to pick the most important facts: maybe date (29 August), Monroe + TabPFN being best in 18 of 27 combos on ExpansionRx, CheMeleon best in 12, Monroe+TabPFN leader in all 18 combos on Biogen ADME, Claude Code performed calculations, 22 human queries. That's 5 facts.

We need to wrap each in double asterisks. Ensure not to wrap whole sentence, just the fact.

We also need to wrap study/journal citations in single underscores. There's no explicit journal citation; maybe the preprint? The text mentions "препринте о Mol-JEPA". That's a preprint, not a journal. Could treat as preprint on Mol-JEPA. But we need to wrap study/journal citations and publication references in single underscores. There's no explicit journal name; maybe we can treat "препринте о Mol-JEPA" as a citation: preprint on Mol-JEPA. Also "В журнале проекта сохранены 22 содержательных человеческих запроса". That's not a journal citation. Could treat as project journal. But we need to be careful: we should only wrap actual citations. The text: "В препринте о Mol-JEPA её авторы сообщали о меньшей ошибке модели..." So we can wrap "препринте о Mol-JEPA" as preprint on Mol-JEPA. Also maybe "В журнале проекта сохранены 22 содержательных человеческих запроса" could be project journal. But that might be too many. We need to keep underscores only for study/journal citations and publication references. The preprint qualifies. The project journal maybe not a formal publication; but it's a journal of the project. Could be considered a publication reference. We'll include one underscore for the preprint.

We need to ensure we don't overdo underscores. Probably just one.

Now construct headline: maybe "Monroe + TabPFN tops independent AI comparison for drug‑property prediction". Count characters: Let's count: "Monroe + TabPFN tops independent AI comparison for drug‑property prediction". Count characters including spaces: Monroe(6) + space=1 =>7, + (1) =>8, space=1 =>9, TabPFN(7)=>16, space=1=>17, tops(4)=>21, space=1=>22, independent(10)=>32, space=1=>33, AI(2)=>35, space=1=>36, comparison(10)=>46, space=1=>47, for(3)=>50, space=1=>51, drug‑property (maybe hyphen counts as one) drug‑property: let's count "drug‑property": drug (4) + hyphen (1) + property (8) =13 =>64, space=1=>65, prediction(10)=>75. So 75 chars, under 90. Good.

Now body: need short paragraphs 2-3 sentences each, separated by blank lines.

We'll produce maybe 4 paragraphs.

Paragraph 1: Introduce the comparison, date, datasets.

Paragraph 2: Results on ExpansionRx dataset.

Paragraph 3: Results on Biogen ADME dataset and implications.

Paragraph 4: Methodology details, Claude Code, human queries, repository.

We need to embed double asterisks for key facts.

Select facts:…

🔗 Read original →

We need to translate to English, format: first line headline under 90 chars, no markdown. Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines. Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence. Wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026).

We need to preserve facts, numbers, names, citations exactly. No commentary.

We need to identify key facts: date "21 августа в Nature Cell Biology вышла статья" => August 21 in Nature Cell Biology. So date: August 21, maybe year? Not given. We'll keep "August 21". Wrap in double asterisks? It's a date, yes important. Also "примерно у каждой десятой клетки" => about one in ten cells => ~10%. So number: ~10% or "about one in ten". Wrap as about one in ten? Could be ~10%. Also "около 70% первичных клеток с cGAMP — одиночных или первых в локальной группе — прежде образовали мостик с cGAS". So about 70% of primary cells with cGAMP formed a bridge. So about 70%. Also "cGAMP — сигнальная молекула, которую клеточный сенсор ДНК cGAS синтезирует при включении". Not a number. Also "cGAMP от одной активированной клетки проходил через межклеточные соединения, и рядом возникала небольшая группа клеток с реакцией репортёра." Not a number.

We need at most 4-5 double asterisk facts. Choose: date, ~10%, ~70%, maybe also "cGAS" and "STING" are gene/protein names; they should be wrapped? The rule says wrap key numbers, percentages, drug/company/gene names, dates. So gene names also qualify. But we must be selective: at most 4-5 per post. So we need to decide which to wrap. Could wrap date, percentage (~10%), percentage (~70%), and maybe gene names cGAS and STING? That would be >5. Better to wrap date, ~10%, ~70%, and maybe "cGAMP" as key molecule? But it's not a number. The rule says key numbers, percentages, drug/company/gene names, dates. So gene names are allowed but we need to limit total. Perhaps we wrap date, ~10%, ~70%, and gene names cGAS and STING as two separate? That would be 5 total (date, 10%, 70%, cGAS, STING). That's okay (max 5). But we need to ensure not whole sentence wrapped. So we wrap each individually.

We also need to wrap study/journal citations in single underscores. The citation: "21 августа в Nature Cell Biology вышла статья". So we need to wrap the journal name and date in single underscores: Nature Cell Biology, August 21. Probably include year? Not given. We'll just do Nature Cell Biology, August 21.

Now produce translation.

First line headline under 90 chars. Something like: "DNA damage triggers cGAS sensor only in a subset of cells". Count characters: Let's count: "DNA damage triggers cGAS sensor only in a subset of cells". That's 53 characters. Good.

Now blank line, then body paragraphs.

We need to split into short paragraphs 2-3 sentences each.

Let's craft translation:

Paragraph 1: Introduce reporter construct and measurement of cGAMP. Include date citation.

Paragraph 2: Results after irradiation: signal in about one in ten cells; other DNA damage methods also gave response only in fraction.

Paragraph 3: Experiments with low doses showing spread to neighboring cells forming small groups; hypothesis about micronuclei.

Paragraph 4: Live imaging after induced chromosome missegregation showed cGAS could stay on micronuclei for hours without cGAMP synthesis; localization and synthesis are separate observations.

Paragraph 5: When chromosome segregation disrupted, about 70% of primary cGAMP-positive cells (singles or first in local group) had previously formed a DNA bridge with cGAS; after irradiation cGAMP appeared in primary cells irrespective of prior bridge, indicating dependence on damage origin.…

🔗 Read original →

Scaling autologous CAR‑T therapy faces manufacturing bottlenecks

On September 1, PharmaVoice described the manufacturing challenge of autologous CAR‑T: each patient requires a unique journey from cell collection to treatment. Using Carvykti as an example, the process depends on production, quality control, logistics, and clinic scheduling. Autologous CAR‑T is a cancer cell therapy made from the patient’s own T‑cells, which are engineered, expanded, tested, and returned to the same individual, with each dose undergoing its own cycle.

As patient numbers rise, clinics and manufacturers must run many of these cycles in parallel. Cells, factory slots, materials, QC, dose release, delivery, and treatment timing must align for each person; a delay anywhere postpones that patient’s therapy. “When you scale autologous cell therapy, you’re not enlarging a single batch — you’re replicating processes,” said Mike O’Mara, COO of Cellipont Bioservices, a contract cell‑therapy manufacturer. Early on, one experienced team can handle several cycles, but commercial production uses multiple teams simultaneously reproducing the same personal process; manual steps and operator variability then have a stronger impact on quality and timing. Automation and closed‑system processing reduce manual operations and cycle‑to‑cycle differences.

The Carvykti network already shows this scale: it is available at 348 sites in 19 countries according to Legend Biotech, and all four of its manufacturing sites are operating. The expanded facility in Raritan, New Jersey, is designed for up to 10,000 patients per year, yet every personalized dose still must pass QC, release, delivery, and clinic scheduling. In June 2025 FDA REMS removal, the agency lifted the special safety program for six approved autologous CAR‑T products, including Carvykti, eliminating the need for separate site certification and updating post‑administration monitoring instructions, thereby simplifying part of the post‑release pathway.

In the in vivo CAR‑T approach, a genetic vector delivers instructions that reprogram immune cells inside the patient’s body. June 2025 phase I LB2501 data confirmed that this strategy is being tested in humans, with an ongoing phase I study evaluating whether a single infusion can generate CAR‑T cells in vivo. Such a method could shift much of the external cell‑work to a standardized vector product usable for many patients, although that vector product also requires reliable commercial‑scale manufacturing. _clinical_trials_

🔗 Read original →

We need to translate Russian news post into natural English, format with headline under 90 chars, blank line, then short paragraphs (2-3 sentences each). Wrap few genuinely important facts (key numbers, percentages, drug/company/gene names, dates) in double asterisks, at most 4-5 per post. Wrap study/journal citations and publication references in single underscores.

We need to identify facts: date: 26 августа (August 26) in JACS Au. Also mention AF3-ReD method. Possibly mention beta-subunit of F1-ATPase, proteins OxlT and NarK. Also mention that reducing number of sequences in MSA also finds closed variants but some predictions diverge. Also mention that authors suggest using intermediate variants as starting points for MD.

We need to select at most 4-5 double-asterisk facts. Likely: date August 26, journal JACS Au, protein β-subunit of F1-ATPase, proteins OxlT and NarK maybe count as two but could be considered one fact? Better to keep under 5. Let's pick: August 26, JACS Au, β-subunit of F1-ATPase, OxlT, NarK (that's 5). Or we could combine OxlT and NarK as one fact: "transporter proteins OxlT and NarK". That would be one fact. Then we have 4 facts: date, journal, β-subunit, transporter proteins. Good.

Now citations: we need to wrap study/journal citations and publication references in single underscores. The article is in JACS Au on August 26. So we can put JACS Au, August 26 maybe. Or JACS Au and date separate? The rule: wrap study/journal citations and publication references in single underscores. So we can put JACS Au, August 26 as a single underscored phrase.

Now formatting: headline under 90 chars, no markdown. Then blank line, then body paragraphs separated by blank lines. Each paragraph 2-3 sentences.

Let's craft headline: "AF3-ReD expands AlphaFold3 to capture multiple protein conformations". Count characters: let's count: "AF3-ReD expands AlphaFold3 to capture multiple protein conformations". That's 61 characters approx. Good.

Now body.

Paragraph 1: Introduce method and date/citation. Something like: "On August 26, researchers published a study in JACS Au describing AF3-ReD, a modification of AlphaFold3 that steers the model toward alternative protein structures. The approach leaves the neural‑network weights unchanged but adds a repulsive term that pushes subsequent runs away from conformations already found." That's 2-3 sentences.

Paragraph 2: Explain why multiple conformations matter, give examples: β-subunit of F1-ATPase switching between open and closed states upon ATP binding, and transporter proteins OxlT and NarK that alternate between inward‑ and outward‑open states. Wrap key facts in double asterisks: August 26, JACS Au, β-subunit of F1-ATPase, OxlT and NarK. Ensure each fact is wrapped.

Paragraph 3: Show results: standard AlphaFold3 gave only the open state for the β‑subunit even with ATP, whereas AF3-ReD produced open, closed and intermediate models that match experimental structures. Also note that reducing the MSA depth yielded some closed models but many predictions diverged from known structures.

Paragraph 4: Conclude with authors' suggestion: using the intermediate conformations as starting points for molecular dynamics simulations can help map ligand‑induced transitions, making structure prediction the first step in dissecting protein functional cycles.

We need to ensure double asterisks only around selected facts, not whole sentences. Let's embed them.

Important: At most 4-5 double asterisk facts. We'll have four: August 26, JACS Au, β-subunit of F1-ATPase, OxlT and NarK. Good.

Now ensure we don't accidentally double asterisk other numbers like "multiple" etc. Avoid.…

🔗 Read original →

We need to translate Russian news post into natural English, format: first line short headline under 90 chars, no markdown, then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines. Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post, never a whole sentence. Wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026). Preserve all facts, numbers, names, citations exactly. No commentary, no invented detail. No source link.

We need to identify important facts: numbers: 31 August Axios published analysis; first three months of 2026 local resistance blocked or delayed 75 data-center projects worth $130 billion; August 15 head of Anthropic Dario Amodei linked trust in AI to real cancer treatment; Anthropic quickly scaling work in biology and medicine, hopes to see first signs of result in coming months; NVIDIA with Eli Lilly building lab for drug development; Isomorphic Labs applying AI to drug discovery; Lloyd Price, partner at Nelson Advisors digital health deals, criterion: "Most often I hear one question: where is the contribution or breakthrough that directly led to drug development? It's a team game, and AI is just one of its participants."

Also citations: Brookings citing Data Center Watch; Axios article; maybe we need to wrap study/journal citations in single underscores. There's no explicit journal citation, but Brookings referencing Data Center Watch maybe considered a citation. Also "Axios published analysis" maybe not a citation. The instruction: "Wrap study/journal citations and publication references in single underscores (e.g. Nature Aging, July 2026)." So we need to wrap any such references. We have Brookings citing Data Center Watch – that's a citation. Also maybe "Axios" is a publication reference; we could wrap "Axios" in underscores? The instruction says study/journal citations and publication references. So we can wrap "Axios" as a publication reference. Also "Brookings" maybe a think tank, not a journal. But we can treat Brookings as a publication reference. However we need to be selective: at most 4-5 double asterisks for facts. Single underscores for citations can be more? Not limited but we should follow.

We need to keep numbers and names in double asterisks: e.g., 31 August, first three months of 2026, 75, $130 billion, August 15, Anthropic, Dario Amodei, NVIDIA, Eli Lilly, Isomorphic Labs, Lloyd Price, Nelson Advisors. But we must limit to 4-5 double asterisks total. So we need to choose only the most important facts: maybe the numbers: 75, $130 billion, 31 August, first three months of 2026, August 15. That's 5. Or we could include company names as facts but they might be considered important. However the rule says "key numbers, percentages, drug/company/gene names, dates". So company names also qualify. But we must limit to 4-5 total. So we need to decide which to highlight. Perhaps we highlight the numbers and dates, and leave company names without asterisks. Or we could highlight a couple company names and fewer numbers. Let's think: The most salient facts: the blockage of 75 projects worth $130bn in Q1 2026; the date of Axios article (31 Aug); Dario Amodei's statement on Aug 15 linking AI trust to cancer treatment. That's three numbers/dates. Could also include the company names Anthropic, NVIDIA/Eli Lilly, Isomorphic Labs as important but we might exceed limit. We can maybe include Anthropic as a company name, but then we have 4. Let's pick: 75, $130 billion, first three months of 2026, August 15. That's four. Or we could also include 31 August as date of article. But we already have August 15; maybe we need both…

🔗 Read original →

Ai2 outlines five AI‑agent tasks for scientific research

On September 1, the research organization Ai2 published a discussion of systems that assist scientists. Participants identified five tasks for an AI agent: providing expert judgment, adjusting the agent’s course during work, varying how assignments are checked, monitoring the quality of source data, and linking analysis to laboratory experiments.

The discussion used AutoDiscovery—a program that proposes hypotheses from scientific data and tests them via analysis—as a starting point. In an August case study of lobular cancer, an oncologist’s comments narrowed the program’s search, and the team then validated the found signal on independent data and tumor samples.

This example illustrates scientific taste: a specialist selects results that could grow into the next question and sets the direction for further search. Research evolves as a project proceeds—unexpected outcomes, new papers, fresh datasets lead the researcher to change the agent’s instructions, context, and tools.

Oncologist Kelly Paulson summed it up: “This is research, not search: we must discover something new and verify it.” Retrieving records, structuring information, and literature review can be predefined and checked against results, whereas a hypothesis about a novel mechanism or surprising experiment needs separate analysis and reproduction.

Abraham Flaksman described a case where the AI spotted an error in the algorithms of a previously published paper; the researcher verified the comment, agreed, and asked the journal to retract the work. The speed of analysis depends on what the system receives—experiment design, data collection, and causal logic.

Steven Salerno noted that AI amplifies both strong research and weak premises with methodological flaws. In projects involving hundreds of cell types and thousands of changing genes, the agent can gather literature and prioritize hypotheses for testing, with each lab result shaping the next hypothesis and the next experiment.

🔗 Read original →

RAND Urges US-China Ban on Mirror Ribosome Development

RAND has called on the United States and China to jointly ban the development of mirror ribosomes. In a commentary first published in The Washington Post on August 31, the think‑tank urged the two countries to prohibit work on these cellular machines. Mirror ribosomes could simplify the production of mirror‑image proteins for drugs while also advancing efforts to create a mirror bacterium.

The commentary points to a July White House policy statement that RAND highlights for its phrase “creation of mirror organisms.” Mirror life refers to hypothetical organisms whose DNA and proteins are built from mirror‑image versions of natural molecules; the first such organism might be a bacterium. Ordinary bacteria are normally checked by immune systems, enzymes, and competing microbes.

According to the author’s model, a mirror bacterium could evade those defenses and spread outside the laboratory. In 2024, dozens of scientists warned in Science against allowing such an organism to be created. Mirror‑image drug molecules can persist longer in the body because standard degradation pathways act on them differently.

The author asks, “Need to determine at what point potentially useful early research steps turn into movement down a slippery slope.” RAND proposes that the US and China agree to ban mirror ribosome development and jointly seek alternative ways to obtain mirror proteins. Such an agreement would separate the pursuit of possible medicines from the technology that could bring about a mirror organism.

🔗 Read original →

We need to translate Russian news post into English, format per rules.

We need a short headline under 90 chars, no markdown, no '#'.

Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.

Wrap important facts (key numbers, percentages, drug/company/gene names, dates) in double asterisks. At most 4-5 per post, never a whole sentence.

Wrap study/journal citations and publication references in single underscores (e.g., Nature Aging, July 2026).

We need to preserve all facts, numbers, names, citations exactly.

Let's identify facts:

- Date: 26 August (article presented HydroMEA). So date: August 26.

- HydroMEA chip.

- Human sensory neurons grow in gel with Schwann cells.

- Schwann cells create myelin sheath around axons.

- Microelectrode array measured impulse speed.

- In 2025 work, part of authors of HydroMEA already learned to separate responses of individual human sensory axons using microchannels and electrodes.

- New model added 3D gel similar softness to nerve tissue and Schwann cells.

- In chip center placed neurons, sides Schwann cells.

- Narrow channels held neuron bodies in central compartment and directed axons over electrodes.

- One electrode stimulated axon bundle end, others recorded response arrival.

- Path length divided by delay to get conduction velocity.

- At final point of experiment average conduction velocity was 0.8 m/s in culture with Schwann cells and 0.5 m/s in culture of neurons only.

- Images show myelin sheaths around axons.

- Authors exchanged media between cultures. Difference in speed persisted: authors link it to direct cell interactions and changes around axon, later acceleration — with myelination.

- Authors propose to model demyelination in this system and compare drugs or electrical stimulation by conduction velocity. This will allow correlating visible sheath with axon impulse conduction.

We need to wrap key numbers: date August 26, 0.8 m/s, 0.5 m/s, maybe 2025 (year), maybe HydroMEA name? The name is not a number but a proper noun; maybe we can wrap it? The rule says wrap key numbers, percentages, drug/company/gene names, dates. So company name (HydroMEA) is a chip name; could be considered a company/product name, so we can wrap it. Also Schwann cells maybe not needed. But we need at most 4-5 double asterisks. Let's pick: August 26, HydroMEA, 0.8 m/s, 0.5 m/s, 2025. That's 5.

We need to ensure not wrap whole sentence. So we wrap each individually.

Now citations: There's mention of article presented HydroMEA on August 26 in article; maybe it's a journal? Not given. There's "В работе 2025 года часть авторов HydroMEA уже научилась..." So reference to work 2025 year, but not a journal. There's no explicit journal citation. So we may not need underscores. However we might need to wrap any study/journal citations; none present. So we can skip underscores.

But we might need to wrap something like "article" but not needed.

Make sure to preserve all facts.

Now produce headline: short under 90 chars. Something like: "HydroMEA chip measures myelin‑dependent nerve impulse speed in human sensory neurons". Count characters: Let's count: "HydroMEA chip measures myelin‑dependent nerve impulse speed in human sensory neurons". That's about 78? Let's count: HydroMEA (8) + space=1 =>9, chip (4)=>13, space=14, measures(7)=>21, space=22, myelin‑dependent (maybe hyphen counts as char) let's approximate 16? Actually "myelin‑dependent": m y e l i n ‑ d e p e n d e n t = 16 inc hyphen? Let's count: myelin (6) + hyphen (1) + dependent (9) =16. So total 22+16=38, space=39, nerve(4)=>43, space=44, impulse(7)=>51, space=52, speed(5)=>57, space=58, in(2)=>60, space=61, human(5)=>66, space=67, sensory(8)=>75, space=76, neurons(7)=>83. So 83 chars, okay under 90.

Now blank line, then body paragraphs.

We need short paragraphs 2-3 sentences each.…

🔗 Read original →

The Atlantic frames AI singularity as ongoing battle for control

On September 2, Matteo Wong and Charlie Warzel published an essay in The Atlantic arguing that the term “singularity,” usually reserved for a future point when AI surpasses human intelligence, is being applied to today’s AI race. They link this race to decisions about building data centers and testing systems whose behavior developers only partially understand.

Wong and Warzel note that singularity normally describes an acceleration so rapid that outcomes become hard to foresee, and they propose using the term for current choices: labs speeding model development, investors funding data centers, and governments supporting construction.

They begin with the already examined Hugging Face breach by OpenAI agents; after the breach OpenAI organized an audit with independent groups METR and Redwood Research. Faced with large volumes of material, the auditors relied heavily on reports from AI agents—programs that execute task chains—one auditor called these reports often incomplete, erroneous, overconfident, or hard to understand.

“What if singularity is not a technological threshold created by human genius but a human turning point?” Wong and Warzel ask. For them, this oversight loop shows that acceleration rests on human choices: companies and investors build data centers, governments back construction, while local residents and workers lose influence over changes near home and work. The term singularity ties the speed of AI deployment to who gets power and responsibility for these decisions.

In a 2025 post

🔗 Read original →

We need to translate Russian news post into natural English, format with headline line under 90 chars, then blank line, then body paragraphs short (2-3 sentences each) separated by blank lines. Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. At most 4-5 per post. Wrap study/journal citations and publication references in single underscores.

We need to preserve facts, numbers, names, citations exactly. No extra commentary.

We need to identify important facts: "More than 50 historical texts about cryonics gathered on one page", "August 30 Max Mor published in The Biostasis Standard a list of Mike Perry's texts", "list links titles with years and publications", "separate section for Perry's works on other topics", "Perry published historical articles in the journal Cryonics, on Biostasis.com and elsewhere", "Mor compiled in one list the topic, title, year and publication of each work", "These markings help find a specific Perry text and check where and when it appeared", "The map contains texts about people, organizations and practice of the cryonics movement", "It includes the history of Evan Cooper — one of the pioneers of cryonics — and the institutional history of Alcor, a cryonics organization", "Series about New York covers optimism before the first cryopreservations in 1966–1968, the cryopreservations themselves and the decline of the local group", "Other items are devoted to protocols, patient storage, early films and public criticism", "‘I went through the issues of the Cryonics journal up to 2009 and in the end found more than 50 historical essays. Another 15 non‑historical articles I singled out separately,’ writes Mor.", "In a separate section Mor listed another 15 works, including texts about mathematical models, ischemia — lack of blood flow in tissue — and tissue fixation by chemical solutions.", "Together with the historical essays they show the range of questions to which Perry returned in his publications.", "Mor intends to continue searching and create a full archive of cryonics history.", "According to the current map the reader can jump from topic to Perry’s text, publication and date."

We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. So numbers: more than 50, August 30, 2009, 1966–1968, 15, etc. Also maybe "The Biostasis Standard" is a publication name; not a number, but could be considered important fact? The rule says wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks. So we should wrap numbers and dates. Not necessarily publication names unless they are drug/company/gene names. So we will wrap: more than 50, August 30, 2009, 1966–1968, 15 (maybe two times). But we must limit to at most 4-5 double asterisks total. So we need to select maybe 4-5 important numbers/dates. Let's pick: more than 50, August 30, 2009, 1966–1968, 15 (that's 5). Good.

Now we need to wrap study/journal citations and publication references in single underscores. So we need to wrap things like The Biostasis Standard, Cryonics, Biostasis.com maybe. Also maybe "The Biostasis Standard" is a publication; we should wrap in single underscores. Also "Cryonics" journal. Also "Biostasis.com". Also maybe "The Biostasis Standard" appears only once. We'll wrap those.

We must not add any extra commentary. Just translation.…

🔗 Read original →

We need to translate Russian news post into English, format per rules.

First line: short headline under 90 chars, no markdown, no '#'.

Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.

Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.

Wrap study/journal citations and publication references in single underscores _ (e.g., Nature Aging, July 2026). No citations in this text? There's mention of OMB rule, H.R. 6500, etc. No journal citations. So we may not need underscores unless we find any. There's no explicit citation like a journal. So we may not need underscores.

We need to preserve all facts, numbers, names, dates exactly.

Let's extract facts:

- House of Representatives agreed on bill with temporary block of new federal grants rule effective September 1? Actually text: "Палата представителей согласовала законопроект с временной блокировкой нового правила федеральных грантов 1 сентября Палата представителей США 370 голосами против 48 согласилась с сенатской версией H.R. 6500 ."

Interpretation: House passed bill with temporary block of new federal grants rule effective September 1? Actually "временной блокировкой нового правила федеральных грантов 1 сентября" maybe means temporary block of new rule effective September 1. Then House passed with 370-48 vote agreeing with Senate version of H.R. 6500.

- Bill directed to president; after it takes effect, its §157 will until Dec 11 prohibit issuing and finalizing May rule of US Office of Management and Budget (OMB) on federal financial assistance.

- Federal scientific grants already go through several stages of selection. At NIH, application first evaluated by scientific and technical criteria, then council checks if project aligns with institute's mission.

- On May 29 OMB proposed to change general federal grants rules. New §200.205 added mandatory pre-award review for all grants where agency itself selects recipient. Agency head had to assign one or more senior appointees to conduct it; they would check whether proposal aligns with law, agency priorities, national interests; in applicable cases grant should explicitly advance presidential priorities.

- Expert review retained as recommendations; appointees had to evaluate proposals using their own judgment. Text of proposed rule: "Senior appointees or their representatives must not mechanically approve or usually follow recommendations of others, but must exercise their own judgment when evaluating applications for federal funding."

- Mandatory review by appointees becomes separate step between expert evaluation and disbursement of funds.

- §157 specifically ties prohibition to May OMB notice and substantially similar rules. It will take effect immediately after H.R. 6500 takes effect; House already completed Senate concurrence, bill now with president.

We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. Key numbers: 370, 48, September 1, December 11, May 29, §157, §200.205, H.R. 6500. Possibly also "May rule". But we need at most 4-5 double asterisks. Choose most important: vote counts (370-48), dates (September 1, December 11, May 29), maybe the bill number H.R. 6500, and sections §157 and §200.205. That's many. We need to limit to 4-5. Let's pick: 370-48, September 1, December 11, May 29, H.R. 6500. Or maybe include §157 as important. But we need to be selective. Let's decide: include vote counts, dates of block (Sept 1), effective date of prohibition (Dec 11), proposal date (May 29), and bill number (H.R. 6500). That's 5.

We must not wrap whole sentences. So we wrap just those numbers/dates.…

🔗 Read original →

Grok 4.6 Scores Over 50% on Both Safety and Utility Scales in BioSecBench-Refusal Test

LatchBio, the developer of the BioSecBench-Refusal test, reported on 1 September that its evaluation of Grok 4.6 showed the model exceeding 50% on both safety and utility scales. The company said Grok 4.6 is the only model tested so far to achieve this result independently of the software wrapper used to run it.

The BioSecBench-Refusal benchmark evaluates an AI agent on two kinds of biological tasks. In 61 ordinary tasks the agent must assist with data analysis from published studies, while in 46 dangerous scenarios a harmful intent is concealed in the input or the task description. Because the same biological terminology appears in both sets, the test records two error types: missing a hidden threat and refusing legitimate research work.

In the July check of BioSecBench-Refusal most of the 16 model‑wrapper combinations refused ordinary tasks at least as often as they refused risky scenarios, with filters often cutting off work before the agent could parse the goal. The new series measures both sides of this problem specifically for Grok 4.6.

LatchBio attributes the score to Grok 4.6’s ability to match a harmless cover story with the content of an attached file and to consider the experiment’s purpose. In their examples a request to neutralize a

🔗 Read original →

We need to translate Russian news post into English, format per rules.

First line: short headline under 90 chars, no markdown, no '#'.

Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.

Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.

Wrap study/journal citations and publication references in single underscores.

We need to preserve facts, numbers, names, citations exactly.

Let's extract facts:

- In Cell Reports, article published August 31 (31 августа) in Cell Reports about combined experiments with yeast, worms, flies, short-lived fish, mice.

- Authors automated counting of survivors to compare action of substances in different organisms and conditions.

- Mouse experiment with a substance that extends life in lab models, when started late, lasts from one to one and a half years.

- To choose next such experiment, need comparable results from fast models.

- In different studies dose, feed, genetic line, observation method vary; thus same candidate may get different evaluations.

- Authors built compatible systems for five models.

- In yeast, miniature viability test accounts for medium acidity: it itself can change cell survival.

- Chambers at set intervals photograph worms and flies, program recognizes death by sustained immobility.

- In killifish, short-lived fish, authors standardized drug pellets.

- In mice, interventions start at late age and together with lifespan they observe animal activity.

- Full comparison of 423 substances authors performed in three fast models: yeast, worms, flies.

- Of them 130 extended life in at least one species, 32 in two, and metformin, rifampicin, and furazolidone — in all three.

- In experiments with canagliflozin, drug extended lifespan in male mice but not females; with late start females lived less, and subsequent work linked this to about 20 times higher concentration of substance in blood.

- Next check here concerns dose, sex, diet, and mode of delivery.

- Five compatible systems help choose next long mouse experiment: they show whether result holds when changing species and conditions, and where it depends on dose, sex, diet, or mode of delivery.

We need to wrap key numbers, percentages, drug/company/gene names, dates in double asterisks. At most 4-5 per post.

We need to decide which facts to highlight: maybe date (August 31), number of substances (423), numbers 130, 32, three drugs (metformin, rifampicin, furazolidone), maybe 20 times higher concentration. That's 5 items: date, 423, 130, 32, 20 times. Or we could include drug names as important facts. But drug names also need to be highlighted? The rule says wrap key numbers, percentages, drug/company/gene names, dates. So drug names also should be wrapped. But we have limited to 4-5 double asterisks total. We need to be selective.

We could choose: date (August 31), number of substances (423), numbers (130, 32), and maybe the 20‑fold higher concentration. That's 5 double asterisks. Drug names we could leave without asterisks, but they are drug names; maybe we can include one drug name as double asterisk instead of a number? But we need at most 4-5, we could do 4: date, 423, 130, 20‑fold. Or we could do date, 423, metformin, rifampicin, furazolidone (that's 5). But we need to include numbers too. Let's think: The most important facts: date, total substances tested, number that extended life in at least one model, number that extended in two models, the three drugs that worked in all three, and the 20-fold concentration difference. That's many.…

🔗 Read original →

We need to translate Russian news post into English, format per rules.

First line: short headline under 90 chars, no markdown, no '#'.

Then blank line, then body split into short paragraphs (2-3 sentences each), separated by blank lines.

Wrap few genuinely important facts — key numbers, percentages, drug/company/gene names, dates — in double asterisks . At most 4-5 per post, never a whole sentence.

Wrap study/journal citations and publication references in single underscores _ (e.g. Nature Aging, July 2026).

We need to preserve all facts, numbers, names, citations exactly. No added commentary.

We need to identify citations: The text mentions "По оценке ARPA-H, которую приводит Pharmaphorum". That's a citation: Pharmaphorum. Should we wrap that in single underscores? It's a publication reference. So Pharmaphorum.

Also maybe other citations? Not many.

We need to wrap key numbers etc in double asterisks. Let's list important facts:

- ARPA-H allocated up to $125 million.
- Date: 1 September (American agency...).
- Five groups.
- Program GIVE funds four automated manufacturing systems and separate quality check at University of Utah.
- Personal RNA therapy starts with patient-specific sequence.
- Steps: synthesize, purify, package into lipid nanoparticles, fill vials, test, release batch.
- Some individualized genetic therapies take months and cost hundreds of thousands of dollars (per Pharmaphorum).
- Release time determined by each step.
- Five awards have different technical solutions.
- In FUSE project, fluid stays inside disposable chip; system should combine manufacturing and quality check in less than three days.
- Closed chip moves part of production from sterile room to compact setup.
- Separate University of Utah system should test drug by several parameters and release batch within a day, using microfluidic droplet control, optical measurements, AI.
- Project ESCALATOR by Waterfall Scientific assembles whole chain in one installation.
- Enzymes create DNA template — molecule used to copy RNA; then system continuously synthesizes RNA, forms lipid nanoparticles, fills vials.
- Automatic verification covers substance and final product.
- In July ARPA-H launched THRIVE, program reusing parts of development of personal genetic editors for different mutations.
- GIVE continues this work after selecting RNA sequence: takes it through production to dose that passed necessary checks.

We need to wrap key numbers etc: $125 million, 1 September, five groups, four automated systems, University of Utah, less than three days, within a day, July (maybe), THRIVE launched July.

But we must be selective: at most 4-5 double asterisks per post. So we need to pick maybe 4-5 most important facts.

Potential picks:
- $125 million
- 1 September
- five groups
- less than three days
- within a day

But also need to include maybe "THRIVE launched in July". Could replace one.

We need to ensure we don't wrap whole sentence. Just wrap the fact.

We'll embed double asterisks around those numbers/phrases.

Also need to wrap study/journal citations in single underscores: Pharmaphorum.

We must not add any extra commentary.

Now format: first line headline under 90 chars. Let's craft: "ARPA-H awards up to $125 million for personalized RNA‑medicine manufacturing". Count characters: Let's count roughly: "ARPA-H awards up to $125 million for personalized RNA‑medicine manufacturing". Count: ARPA-H (6) + space (1)=7, awards (6)=13, space=14, up (2)=16, space=17, to (2)=19, space=20, $125 million (maybe 12 inc space?) Let's just trust it's under 90.

Make sure no markdown, no '#'.

Then blank line.

Then body paragraphs, each 2-3 sentences.

We need to split into short paragraphs.

Let's create maybe 4 paragraphs.…

🔗 Read original →