ents treated with DT120. From Definium Therapeutics’ Corporate Presentation , page 36: So what’s the big deal? DT120 could potentially treat major depressive disorder (MDD) and generalized anxiety disorder (GAD). So far they got one jaw-dropping phase 3 trial readout for MDD already out , two more phase 3 trial readouts for GAD to come in early Q3 2026, and in late Q3 2026 (worth mentioning: Definium, named MindMed at the time, was awarded Breakthrough Therapy Designation by the FDA because of how groundbreaking DT120 proved to be for GAD in phase 2), not to mention a phase 3 trial for Post-Traumatic Stress Disorder (PTSD) starting next year. You’re confused? Here’s a quick snapshot from the Corporate Presentation , page 8, with all Definium’s clinical trials + upcoming catalysts for DT120. Yeah, yeah, it all sounds very complicated, doesn’t it? But it isn’t . In layman terms, psychedelics like LSD have been relegated to the fringes of medical science for ages. And not even the fringes, beyond the fringes: they were basically the scourge of the earth, or, as three-letter agencies use to say, a Schedule I drug, on par with heroin. Some years ago Definium Therapeutics and other companies kickstarted a new age of psychedelic medicine — this time based on clinical trials under the guidance of the FDA. And what they discovered was massive. DT120, if all these trial readouts go well — which, of course, they might not, but let’s entertain the possibility everything’s peachy for a moment — is nothing short of a miracle drug . A single, best-in-class, almost-one-and-done formulation with a potential total addressable market (TAM) of 200 billion dollars. And I’m not fabricating numbers, it’s all in black and white on page 36 of Definium’s Corporate Presentation . I promised you math, didn’t I? So here it is. $2 billion revenue opportunity for every 1% of market penetration. Let’s assume trials go well. How big of a market share will DT120 acquire by, say, 2030? Let’s be pessimistic: let’s say 5% . 5% times 2 billion dollars per 1% penetration = 10 billion dollars. What’s the price to sales ratio for an average biotechnology company? According to the dataset of Aswath Damodaran , Professor of Finance at Stern School of Business of New York University, (dataset updated to January 2026), the answer is 7.31 . 7.31 times 10 billion dollars gives us a market cap of 73.1 billion dollars . You remember I told you there are currently 132.74M shares? Let’s be overly pessimistic and assume dilution up to 200M shares — mind you, more dilution is unlikely at this point, given the company has roughly over a billion dollars in cash, which should be enough to commercialize DT120 and get to profitability, but let’s crank caution up to eleven, just to be sure. So, what’s 73.1 billion dollars divided by 200M shares? It’s $365.50 per share. And I remind you, as of today, the company is worth 43.17 dollars a share. You read that right. Over 8x in 4 years , assuming that trial results confirm the greatness the company has got us accustomed to so far. And keep in mind, this is the base case, i.e., Definium gets to only 5% market penetration. Let’s be bold and consider, just for the sake of it, that there is no dilution and DT120 is truly a miracle drug, with a 20% market penetration. In such a case, the stock price would balloon to $2,202.80 per share. Which is 51.03x from today’s price. Many things could go wrong. In no particular order: There’s a non-zero chance

that phase 3 clinical trials do not confirm past results . I think this is unlikely. For starters, check the Corporate Presentation , page 15: The first MDD phase 3 trial for DT120, Emerge, positioned DT120 as best-in-class, i.e., better than any other drug on the market. Isn’t it somewhat safe to assume that the second MDD phase 3 trial for DT120, Ascend, might confirm Emerge’s results? And after the sensational results of DT120 for the treatment of GAD, which were published in JAMA in 2025 , is it all that absurd to assume that the upcoming phase 3 trial results for GAD are going to be… good? I don’t think so. There’s a non-zero chance that, even with great results, the FDA doesn’t support the rescheduling of DT120 . Yes, this could theoretically happen, but it’s highly unlikely. Especially after the White House’s Executive Order to further accelerate psychedelic medicine , which the House codified a few days ago . Plus, this FDA is very supportive of psychedelics and went so far as to give positive guidance on the matter this past July . Whatever you think of the Trump administration, this is probably the best time ever to be a psychedelic-medicine company. That’s it. I might be blind, but I don’t see any other problem. Sure, you might’ve heard from bulls on other psychedelics companies that DT120 gives you a 12-hour trip (which is false, by the way — the average time required for participants to satisfy the study’s structured end-of-session criteria , as reported by Definium , is 5.8 hours) and that long-duration psychedelics are not suited for the current US insurance model. I think this is baseless. First, who cares about the duration of the trip? What matters is how long you stay in remission, not how long the trip is. As of today, Spravato, an esketamine-based treatment for treatment-resistant depression (TRD), requires patients to visit a healthcare setting twice a week for the first four weeks, once a week for the following four weeks, and then once a week or every two weeks during maintenance. That’s a hell of a lot of time! We don’t know how powerful DT120 will prove to be (we need all phase 3 trial readouts to be sure), but it might well be something you take a few times a year, maybe fewer. The question then becomes, is there going to be a sufficient number of clinics to allow for even a 5% market penetration in such a short amount of time? That we don’t know, but in my opinion, it’s safe to assume that DT120 might piggyback on the infrastructure that was and is already being established for Spravato (which is a blockbuster drug, by the way). Not to mention that, if DT120 proves to be as effective as it seems to be, new clinics will be established. My point is, patients can’t wait for some new drug to fulfill the promises that SSRIs and many other therapies have already made. The only question is, will you make money out of it? Please, for the love of God and whatever else you deem sacred, don’t follow me on this trade only because you see big numbers on a screen. I’m already up almost 600% on DFTX, and I’m irresponsibly long. My cost basis is in the single digits. Even if trial results are bad and the stock gets cratered, odds are I will get out unscathed, or even make some money out of it. In other words, you and I are not friends, in fact, you should not take my word for it, or anyone else’s for that matter, when it comes to making money. The market is a bad place. Do your own research. Be cautious. Don’t buy 0DTE

options or anything of the sort. Don’t bet the farm on a single trade. And don’t do anything stupid, either financial or medical. Don’t test these drugs on yourself. Don’t do anything your grandma wouldn’t do. I’m not your financial or medical advisor. I’m not your friend. This is not financial advice, or medical advice. In fact, this is not advice. I’m only saying this for posterity, so if it happens, I can say Fingers crossed. And good luck.

反AI情绪中的“过去崇拜”与法西斯主义逻辑的相似性

近年来，科技界反AI情绪高涨，部分批评者怀念“过去真正的程序员”，认为现代大语言模型（LLM）是行业堕落的象征，用低质量代码和数量至上取代了精神层面的匠心。然而，有分析指出，这种对过去的崇拜和对现代技术的拒绝，其论证结构与法西斯主义经典特征高度吻合。文章引用朱利叶斯·埃沃拉、埃兹拉·庞德和希特勒的言论，发现反AI叙事中常见的“精神堕落”“机器统治”“数量凌驾于质量”等修辞，与法西斯文本如出一辙。尽管AI本身常被指责为法西斯工具，但作者提醒：反AI阵营将世界描绘成传统男子气概的精神力量与堕落资本之间的斗争，同样滑入了极权思维的陷阱。这种“过去崇拜”和“拒绝现代主义”正是乌姆贝托·艾柯定义的“原法西斯主义”核心特征。文章呼吁批判技术时需警惕自身论证中的潜在风险。 #反AI #法西斯主义 #技术批判 #过去崇拜 #意识形态 #科技评论 #文化分析 #LLM

C++ Vs Rust: Which is better for writing AI/ML code with LLMs

We do many conversions of AI models from the reference PyTorch implementation to GGML and C++. The reason is GGML/C++ produces a relatively tiny package we can run almost anywhere. The performance usually matches or exceeds PyTorch with a fraction of the dependencies. Personally I despise C++; can’t stand it. I don’t have to write it anymore though, LLMs do it. I sit above a layer of abstraction and only descend into the code to avoid worst case scenarios in much the same way I would randomly go read the code for a library I depended on before AI. Someone asked if we had considered doing these ports in Rust. I am well aware of Rust, got excited about it in the past, but had not seriously considered it because GGML is written in C++. In recent years I experimented by comparing Rust against Go and it was a slam dunk

victory for Go on the project I was doing. Neither myself, LLMs or the person advocating for Rust seemed to understand how to solve basic problems without tripping over the type system. However I do like to explore options and it had occurred to me already that an experiment between Rust and C++ could be interesting, so having someone ask for it is the push I needed to burn some tokens. I in fact ended up doing a series of A/B experiments, none of which are very scientific, but I want to test realistic scenarios. Creating neat test environment tends to increase in difficulty the more realistic you try to make it. The first experiment was to pit C++ with GGML Vs Rust with Burn. They both had to reach parity, in terms of correctness and performance against the PyTorch implementation of FLOAT (it animates portraits given voice audio to make it appear they are talking). The final experiment was to create a pure CPU only implementation using SIMD libraries in C++ and Rust. Removing the confonders introduced by pitting GGML against Burn. The original suggestion was to use GGML from unsafe blocks in Rust, but I thought that this might put Rust at a disadvantage, nullifying its safety properties. Rightly or wrongly I decided to go with Burn instead and I can say right now Burn is great. Except that is on CPU, it’s terrible on CPU. On GPU though it is competitive with GGML and PyTorch on inference and it does training too. I have published Fable’s and ChatGPT’s analyses in the rust-vs-cpp-analysis repository , but here I’ll give my opinions and highly condensed version of events. I asked Claude to create 3 independent sub-agents; one to do an analysis of the PyTorch implementation, create a parity harness to do layer by layer numerical comparisons and do some Rust research to level the playing field with C++. You see, the thing is, the C++ agent would have a fair more prior art to go on, because we have done a bunch of these conversions and it helps to give access to those prior conversions. I’m not going to withold that info because I want results. The best I can do is do some extra work upfront to try and convert lessons learnt on C++ projects into lessons for a Rust proejct. The other two subagents are the Rust and C++ implementors. They ran until they reached correctness and performance parity with the PyTorch implementation. Initially I thought I had walked into a decisive result proving Rust’s greatness. The Rust agent reported about 50% of the token usage while achieving the same performance. However it turned out the Rust agent benefited from some work the C++ agent did to debug the parity harness. The primary Fable agent also decided to do the GGML implementation in CUDA and the Rust one in Vulkan. Then blamed me for asking that GGML use CUDA and Burn use Vulkan. I don’t recall asking for that, but in any case it is a confounder, but not a big one as it turns out. Regardless of whether we use Vulkan or CUDA, fused kernels are required in GGML and something similar in Burn to get parity with compiled PyTorch. After some iterations trying to get the Rust and C++ ports to be equivalent, I decided to do something more drastic as a final test. I asked to create two new ports, both in new repositories, but using lessons learned from the previous attempts. This time the ports would be CPU only and use SIMD libraries. I demanded that they be statically compiled and allow cross compiling to ARM. Do zero allocations during

computations and a bunch of other stuff. Again it initally appeared that Rust had achieved a victory, but it turned out that the rust agent simply decided not to create a C API. Possibly because I didn’t explicitly ask for it, but of course there should always be a C API/ABI otherwise only other Rust code can consume the library. The Rust port was also a bit slower, because the agent simply decided to stop with perf optimizations sooner. Maybe Claude doesn’t like optimizing Rust or more likely it’s just the random nature of LLMs coming through. In any case after forcing the agent to complete the C API/ABI and get perf up to parity, the Rust advantage disappeared. Finally I decided to do a couple of code reviews, one with Claude again and another with Codex. The argument in favor of Rust would predict that the Rust code would have less previously undiscovered bugs. Both Rust and C++ were fuzzed from the beginning, but still the review found plenty of potential bugs in both. I asked both Claude and Codex if Rust’s and C++’s tools and features had been used to their fullest to prevent bugs and the answer was no. Not even close. The agent just didn’t decide to do it on the first round. The agent had made some spurious uses of unsafe, but actually most of the uses seemed to be valid. Removing the unnecessary unsafe blocks wouldn’t have made a lot of difference to the analysis. So essentially both languages have more intrinsic or extrinsic features available to combat bugs. Both agents failed to use all of those features and both agents ended using similar amounts of tokens and time. From what I can gather there is no difference to an LLM between Rust and C++. The main difference between these lanauges is that Rust has some features built into the compiler whereas C++ relies on external tools. There are many problems with C++, but they get arbitraged with external tools like fuzzers, the address sanitzer, static anlysers and more ergonomic libraries. Rust is perhaps more prone to solving these issues in the language. In either case it’s not automatic that language features or external tools will solve the problems they can solve. The surprising part is that solutions which are enforced by the language don’t really solve the original problem automatically. Let’s say you solve memory safety with a borrow checker, now you have a new problem which is working around the borrow checker. Except it is not really a new problem, you are still trying to solve the underlying problem, but with new constraints. That isn’t to say the borrow checker can’t help solve the problem, it’s just that it requires effort to do so. The question is does it require less effort than testing the code with the address sanitizer on, using smart pointers, a GC or using a combination. Possibly all of these techniques are mathematically and computationally equivalent when carried out by an LLM, except that it makes a difference when the computation occurs; in production, test time or compile time. Compile time and test time are practically the same it seems, so C++ and Rust are the same. They both move computation from runtime to development. I suppose the fundamental problem is that you have to model the real world task the software is trying to orchestrate. You have to express this model in the programming language’s primitives. The real world, being as varied as it is, means that any general purpose programming language lacks primitives that

abstract or verify an arbitrary model. Even if the language does contain primitives to help with a large part of your model. You have to know how to use those features, which is not free. Especially when you view it through the lens of agentic codimg. You can put a token count on finding the correct primitives to express the model in. Intuitively I think one language is better, but I haven’t been able to disprove the null hypothesis. That’s most likely because there are things that matter a lot more than whether you write your loops over arrays in C++ or Rust. Now both these languages are created for humans and there is the question of what happens if you systematically optimize a language to reduce token usage while completing practical tasks. It could be a deadend where the language always collapses into degenerate solutions or it could lead to something much better. My feeling is you’d hit diminishing returns quite quickly because the language would need to expand to encode solutions for an increasing number of problems, making recall the bottleneck. However the initial gains could be large.

长鑫科技上市首日暴涨471%，野村看高至116元

7月27日，国产DRAM龙头长鑫科技登陆科创板，开盘即大涨471.59%，报49.5元，市值达3.3万亿元，成为A股市值最大公司。此次IPO发行价8.66元，募资约579亿元，为科创板史上最大IPO。同日，野村证券发布首份覆盖研报，给予“买入”评级，目标价116元，较发行价上涨空间超12倍，对应市值7.76万亿元。野村认为，全球AI驱动内存需求呈指数级增长，而供应端受制于产能扩张瓶颈，供不应求将成为常态。长鑫科技作为中国最大DRAM制造商，市场份额将从当前约10%提升至2028年底的18%，逐步逼近美光。技术层面，公司计划2026年主流制程达16-17nm，2027年量产HBM3。野村预计2026-2028年营收和净利润复合增长率分别达63%和74%，目标价基于20倍2028年预期市盈率。但报告也提示地缘政治为主要风险。 #长鑫科技 #DRAM #IPO #芯片 #AI存储 #野村证券 #科创板

英伟达通过推出NVIDIA PhysicsNeMo和CUDA-X库扩展了NVIDIA Agent Toolkit，以彻底改变全球工程、设计和建造的方式

Operations too frequent. Try again later Page not found, please try again later. Take me home ! : 403 Services by Moomoo Technologies Inc.