Chat GPT
The Bitter Lesson
We could have been talking to our desktop computers in English since the 90s!
"Somebody got one of the small versions of Llama to run on Windows 98…”
“We could've been talking to our computers in English for the last 30 years"
- Marc Andreessen
Correct.
The hardware already existed, for decades.
What stopped us?
Extreme aversion to investing money into training much larger AI models.
No one was willing to invest the many millions needed to train an AI model of this size.
In fact, even a decade later in 2011, people were still hardly willing to spend more than TEN DOLLARS on electricity costs to train a state-of-the-art model, e.g. the AlexNet image model
Many truly under-estimate how unwilling to people have been to spend money on AI training, until very recently
And this wasn’t unrealized, many of us had screamed this for decades.
No one cared.
Incredible testiment to man’s unwillingness to invest in certain critical areas of future tech.
— happens in AI, advanced market mechanisms, proof systems, and a few other similar areas, that are unquestionably the future.
We could have been talking to our desktop computers in English since the 90s
Bitter Lesson
"Somebody got one of the small versions of Llama to run on Windows 98…”
“We could've been talking to our computers in English for the last 30 years"
- Marc Andreessen
Correct.
The hardware already existed, for decades.
What stopped us?
Extreme aversion to investing money into training much larger AI models.
No one was willing to invest the many millions needed to train an AI model of this size.
In fact, even a decade later in 2011, people were still hardly willing to spend more than TEN DOLLARS on electricity costs to train a state-of-the-art model, e.g. the AlexNet image model
Many truly under-estimate how unwilling to people have been to spend money on AI training, until very recently
And this wasn’t unrealized, many of us had screamed this for decades.
No one cared.
Incredible testiment to man’s unwillingness to invest in certain critical areas of future tech.
— happens in AI, advanced market mechanisms, proof systems, and a few other similar areas, that are unquestionably the future.
We could have been talking to our desktop computers in English since the 90s
Bitter Lesson
This media is not supported in your browser
VIEW IN TELEGRAM
We could've been talking to our computers in English for the last 30 years
35.9 tok/sec on a 26 year old Windows 98 Intel Pentium II CPU, with 128MB RAM
Using a 260K LLM with Llama-architecture
35.9 tok/sec on a 26 year old Windows 98 Intel Pentium II CPU, with 128MB RAM
Using a 260K LLM with Llama-architecture
This media is not supported in your browser
VIEW IN TELEGRAM
We could've been talking to our computers in English for the last 30 years
Somebody got one of the small versions of Llama to run on Windows 98…
We could've been talking to our computers in English for the last 30 years
- Marc Andreessen
Somebody got one of the small versions of Llama to run on Windows 98…
We could've been talking to our computers in English for the last 30 years
- Marc Andreessen
OpenAI Introduces “Operator”
Today we’re releasing Operator(opens in a new window), an agent that can go to the web to perform tasks for you. Using its own browser, it can look at a webpage and interact with it by typing, clicking, and scrolling.
To ensure a safe and iterative rollout, we are starting small. Starting today, Operator is available to Pro users in the U.S. at operator.chatgpt.com
Operator is powered by a new model called Computer-Using Agent (CUA). Combining GPT-4o's vision capabilities with advanced reasoning through reinforcement learning, CUA is trained to interact with graphical user interfaces (GUIs)—the buttons, menus, and text fields people see on a screen.
Operator can “see” (through screenshots) and “interact” (using all the actions a mouse and keyboard allow) with a browser, enabling it to take action on the web without requiring custom API integrations.
To get started, simply describe the task you’d like done and Operator can handle the rest. Users can choose to take over control of the remote browser at any point, and Operator is trained to proactively ask the user to take over for tasks that require login, payment details, or when solving CAPTCHAs.
OpenAI Announcement
Today we’re releasing Operator(opens in a new window), an agent that can go to the web to perform tasks for you. Using its own browser, it can look at a webpage and interact with it by typing, clicking, and scrolling.
To ensure a safe and iterative rollout, we are starting small. Starting today, Operator is available to Pro users in the U.S. at operator.chatgpt.com
Operator is powered by a new model called Computer-Using Agent (CUA). Combining GPT-4o's vision capabilities with advanced reasoning through reinforcement learning, CUA is trained to interact with graphical user interfaces (GUIs)—the buttons, menus, and text fields people see on a screen.
Operator can “see” (through screenshots) and “interact” (using all the actions a mouse and keyboard allow) with a browser, enabling it to take action on the web without requiring custom API integrations.
To get started, simply describe the task you’d like done and Operator can handle the rest. Users can choose to take over control of the remote browser at any point, and Operator is trained to proactively ask the user to take over for tasks that require login, payment details, or when solving CAPTCHAs.
OpenAI Announcement
This media is not supported in your browser
VIEW IN TELEGRAM
OpenAI’s Operator reading through hotel reviews on Tripadvisor to find the best hotel sauna in Stockholm
This media is not supported in your browser
VIEW IN TELEGRAM
A more realistic Hal 9000
Forwarded from DoomPosting
OpenAI releases o3 and o4-mini
Highlights,
(1) Reinforcement learning training — a DECADE after DeepMind thoroughly showed that the RL approach is the way to go, and AFTER OpenAI getting thoroughly wrecked by DeepSeek who heavily used the RL approach — OpenAI FINALLY gets off of their retarded wordcel asses, and starts to shift in the rotator direction, finally making much heavier use of RL training. OpenAI also pretends to be surprised in further confirming the ~unbounded RL training scaling laws that DeepMind thoroughly confirmed a decade ago.
(2) Real thinking with images — OpenAI FINALLY making images more of a first-class medium, “For the first time, these models can integrate images directly into their chain of thought. They don’t just see an image—they think with it.”
(3) Codex CLI — apparently a new AI agent, through which their latest models were trained to be good at using the terminal - huge for giving the AI real agency, instead of constantly needing help from humans for mundane things
(4) Agentic tool use — OpenAI FINALLY gives their top models full access to all their internal “tools”, and heavily trains the models with RL to best use the tools. Big question hear is WHY THE F&$& did they wait years before finally giving their top models full tool access. OpenAI had been weirdly restricting their top models to different arbitrary subsets of the tools up to now.
^^ OpenAI has finally done the few things critical needed to enable powerful agentic AIs
New huge AI wave finally inbound?
OpenAI Announcement
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
Highlights,
(1) Reinforcement learning training — a DECADE after DeepMind thoroughly showed that the RL approach is the way to go, and AFTER OpenAI getting thoroughly wrecked by DeepSeek who heavily used the RL approach — OpenAI FINALLY gets off of their retarded wordcel asses, and starts to shift in the rotator direction, finally making much heavier use of RL training. OpenAI also pretends to be surprised in further confirming the ~unbounded RL training scaling laws that DeepMind thoroughly confirmed a decade ago.
(2) Real thinking with images — OpenAI FINALLY making images more of a first-class medium, “For the first time, these models can integrate images directly into their chain of thought. They don’t just see an image—they think with it.”
(3) Codex CLI — apparently a new AI agent, through which their latest models were trained to be good at using the terminal - huge for giving the AI real agency, instead of constantly needing help from humans for mundane things
(4) Agentic tool use — OpenAI FINALLY gives their top models full access to all their internal “tools”, and heavily trains the models with RL to best use the tools. Big question hear is WHY THE F&$& did they wait years before finally giving their top models full tool access. OpenAI had been weirdly restricting their top models to different arbitrary subsets of the tools up to now.
^^ OpenAI has finally done the few things critical needed to enable powerful agentic AIs
New huge AI wave finally inbound?
OpenAI Announcement
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
Forwarded from DoomPosting
OpenAI Releases Codex
Codex CLI is built for developers who already live in the terminal and want ChatGPT‑level reasoning plus the power to actually run code, manipulate files, and iterate – all under version control. In short, it’s chat‑driven development that understands and executes your repo.
• Zero setup — bring your OpenAI API key and it just works!
• Full auto-approval, while safe + secure by running network-disabled and directory-sandboxed
• Multimodal — pass in screenshots or diagrams to implement features
• Fully open-source
Usage examples:
Interactive REPL:
Initial prompt for interactive REPL:
Non‑interactive "quiet mode”:
OpenAI Codex
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶
Codex CLI is built for developers who already live in the terminal and want ChatGPT‑level reasoning plus the power to actually run code, manipulate files, and iterate – all under version control. In short, it’s chat‑driven development that understands and executes your repo.
• Zero setup — bring your OpenAI API key and it just works!
• Full auto-approval, while safe + secure by running network-disabled and directory-sandboxed
• Multimodal — pass in screenshots or diagrams to implement features
• Fully open-source
Usage examples:
Interactive REPL:
codexInitial prompt for interactive REPL:
codex "fix lint errors”Non‑interactive "quiet mode”:
codex -q --json "explain utils.ts”OpenAI Codex
🄳🄾🄾🄼🄿🄾🅂🅃🄸🄽🄶