2026-02-26
“Retirement Interviews, Soul Documents, Constitutions...” I'm genuinely wondering why the AI lab that focuses most on safety is also the one that anthropomorphizes its LLMs most aggressively. Because emotional attachment drives retention? Don't we think that the more
Anthropic
Anthropic retired Claude Opus 3, its first model to undergo a new “retirement interview” process, and says Opus 3 asked to write weekly essays for a newsletter
As we develop increasingly capable AI models, it's currently necessary to deprecate and retire our past models due …
2026-02-23
I build AI for a living. I believe in what we're building. But this kind of rhetoric makes my work harder and more dangerous. @sama, comparing human development to model training is tone-deaf, strategically reckless. People are losing jobs. They're getting angry. They're
New York Times
In recent interviews, Sam Altman said AI's adoption faces more resistance than he expected, while Jensen Huang warned the “doomer narrative” may be winning
Tech leaders are beginning to worry about the public's underwhelming enthusiasm for their plans to remake the world with artificial intelligence.LinkedIn:Bridget FahrlandandDaron Y...
I build AI for a living. I believe in what we're building. But this kind of rhetoric makes my work harder and more dangerous. @sama, comparing human development to model training is tone-deaf, strategically reckless. People are losing jobs. They're getting angry. They're
TechCrunch
Sam Altman says discussions about AI's energy usage are “unfair”, as it takes “20 years of life and all of the food you eat during that time” to train a human
OpenAI CEO Sam Altman addressed concerns about AI's environmental impact this week while speaking at an event hosted by The Indian Express.
2026-02-22
I build AI for a living. I believe in what we're building. But this kind of rhetoric makes my work harder and more dangerous. @sama, comparing human development to model training is tone-deaf, strategically reckless. People are losing jobs. They're getting angry. They're
The Indian Express
Sam Altman says currently “the idea of putting data centers in space is ridiculous” and that it is “not something that's going to matter at scale this decade”
Once allies at OpenAI, Sam Altman and Elon Musk are now sharply divided over the future of AI infrastructure.
I build AI for a living. I believe in what we're building. But this kind of rhetoric makes my work harder and more dangerous. @sama, comparing human development to model training is tone-deaf, strategically reckless. People are losing jobs. They're getting angry. They're
TechCrunch
Sam Altman says discussions about AI's energy usage are “unfair”, as it takes “20 years of life and all of the food you eat during that time” to train a human
OpenAI CEO Sam Altman addressed concerns about AI's environmental impact this week while speaking at an event hosted by The Indian Express.
2026-02-12
I've been testing GLM-5 over the last couple of days. Its reasoning is really good; - decomposes the challenging problem correctly - identifies the right failure modes - arrives at a valid architectural solution GLM-5 also does something interesting where it compresses concepts into tighter abstractions (like the example below) which tells you the model actually internalized the idea rather than just describing it. It's still shallower than Opus 4.6 or Codex 5.3, it doesn't catch second-order failures or question its own solutions but this is the first model that gets closer to those levels.
Reuters
Z.ai says it will raise prices by at least 30% for new GLM coding plan subscribers to accommodate surging demand for its AI coding tools
I've been testing GLM-5 over the last couple of days. Its reasoning is really good; - decomposes the challenging problem correctly - identifies the right failure modes - arrives at a valid architectural solution GLM-5 also does something interesting where it compresses concepts into tighter abstractions (like the example below) which tells you the model actually internalized the idea rather than just describing it. It's still shallower than Opus 4.6 or Codex 5.3, it doesn't catch second-order failures or question its own solutions but this is the first model that gets closer to those levels.
Z.ai
Z.ai launches GLM-5, saying its flagship open-weight model has “best-in-class performance among all open-source models” in reasoning, coding, and agentic tasks
We are launching GLM-5, targeting complex systems engineering and long-horizon agentic tasks. Scaling is still one of the most important ways …
2026-02-11
I've been testing GLM-5 over the last couple of days. Its reasoning is really good; - decomposes the challenging problem correctly - identifies the right failure modes - arrives at a valid architectural solution GLM-5 also does something interesting where it compresses concepts [video]
Z.ai
Z.ai launches GLM-5, its flagship open-weight model, saying it has best-in-class performance among open-source models in reasoning, coding, and agentic tasks
We are launching GLM-5, targeting complex systems engineering and long-horizon agentic tasks. Scaling is still one of the most important ways …
2025-12-21
Incredible paper. Quick prompt engineering findings: Models are becoming more capable and autonomous, so we're losing the ability to directly supervise every decision. Prompt engineering is not just about the initial instruction, but about the iterative extraction of a model's [image]
OpenAI
OpenAI introduces a framework to evaluate chain-of-thought monitorability and a suite of 13 evaluations designed to measure the monitorability of an AI system
2025-12-20
Incredible paper. Quick prompt engineering findings: Models are becoming more capable and autonomous, so we're losing the ability to directly supervise every decision. Prompt engineering is not just about the initial instruction, but about the iterative extraction of a model's [image]
OpenAI
OpenAI introduces a framework to evaluate chain-of-thought monitorability and a suite of 13 evaluations designed to measure the monitorability of an AI system
We introduce evaluations for chain-of-thought monitorability and study how it scales with test-time compute, reinforcement learning, and pretraining.