"I'm not so interested in LLMs anymore,” says @ylecun who is investigating World Models: JEPA – Joint Embedding Predictive Architecture. Read these 2 posts:
1) https://
linkedin.com/posts/ashley–
nicholson_im-not-so-interested-in-llms-anymore-activity-7366084551552548866-C8RK?utm_source=share&utm_medium=member_ios&rcm=ACoAAAN5amABNvxwNhChBWp6XqXktRsmEAUmhcg
… 2) https://
linkedin.com/posts/starkbur
ns_the-ai-hype-is-a-dead-man-walking-the-math-activity-7364428630296256514-D_L3
…
LLMS
-
Yann LeCun shifts focus from LLMs to World Models and JEPA
By
–
-
Teaching AI Models Physical Reasoning and Common Sense
By
–
How do you teach an AI model to reason? 🤔
— NVIDIA AI (@NVIDIAAI) 27 août 2025
We are developing a set of tests that coach AI models to understand the physical world and apply common sense.
These tests are used to develop reasoning models such as NVIDIA Cosmos Reason which just topped the physical reasoning… pic.twitter.com/eBVNIbQhZoHow do you teach an AI model to reason? We are developing a set of tests that coach AI models to understand the physical world and apply common sense. These tests are used to develop reasoning models such as NVIDIA Cosmos Reason which just topped the physical reasoning
-
GPT-5 Outperforms Physicians in Clinical Reasoning
By
–
GPT-5 Just Outperformed Physicians in Clinical Reasoning A new study (arXiv: 2508.08224 : https://
arxiv.org/pdf/2508.08224) marks a turning point in medical AI: GPT-5 has surpassed licensed physicians on MedXpertQA, one of the most demanding benchmarks for clinical reasoning. Unlike -
GPT-5 championship at deceptive snow job Olympics
By
–
If there was a Snow Job Olympiad, GPT-5 would be a real champion.
-
Claude Code Web Version Announced
By
–
Claude Code agent will paw its way to the web version too. Similarly to Codex, users will be able to run Claude Code without relying on its CLI version. Research Preview
-
Cerebras Inference Celebrates One Year with 6x Speed Gains
By
–
🎂 Cerebras Inference turns 1! 🚀
— Cerebras (@cerebras) 27 août 2025
Let's break it down:
– 6x faster than when we launched — From @Meta Llama to @Alibaba_Qwen 3 to @OpenAI OSS, models running on Cerebras deliver 𝟯,𝟬𝟬𝟬+ 𝘁𝗼𝗸𝗲𝗻𝘀/𝘀𝗲𝗰
– Our inference powers the best AI natives, global enterprises, and… pic.twitter.com/fVVAGcQY7sCerebras Inference turns 1! Let's break it down: – 6x faster than when we launched — From @Meta Llama to @Alibaba_Qwen 3 to @OpenAI OSS, models running on Cerebras deliver 𝟯,𝟬𝟬𝟬+ 𝘁𝗼𝗸𝗲𝗻𝘀/𝘀𝗲𝗰
– Our inference powers the best AI natives, global enterprises, and -
LLMs Show Promise in Medicine and Education, Need More Investment
By
–
We are underinvesting in areas where research suggests AI could be good enough, right now, to make an impact Controlled studies find that current LLMs can be very useful in medicine & education, especially in disadvantaged areas, but more work is needed to deliver on the promise
-
RL Finetuning: Major Upgrade Over SFT for LLMs
By
–
I just mean long term, imo RL finetuning paradigm is a big upgrade over just SFT (expert imitation) for LLMs at the current stage of development and will continue to grow substantially.
-
Training Data Evolution: Internet Text to Supervised Conversations
By
–
In era of pretraining, what mattered was internet text. You'd primarily want a large, diverse, high quality collection of internet documents to learn from.
— Andrej Karpathy (@karpathy) 27 août 2025
In era of supervised finetuning, it was conversations. Contract workers are hired to create answers for questions, a bit… https://t.co/rR6yYZGgKPIn era of pretraining, what mattered was internet text. You'd primarily want a large, diverse, high quality collection of internet documents to learn from. In era of supervised finetuning, it was conversations. Contract workers are hired to create answers for questions, a bit
-
GPT-5 Performance Matters More Than Minor Flaws
By
–
And people still make fun of the fact that ChatGPT sometimes doesn't count fingers correctly on photos. I don't care if GPT-5 still has teething problems as long as it does a good job in the areas that matter. And when one of the leading experts in his field tells me that GPT-5