explore | Grilled Cheese

What if the real alignment problem isn’t to get AI systems to be aligned with human goals and values, but to get *AI companies* to be so, one being a function of the other? #AIEthics #LLMs

OpenAI has internal, fine-tuned LLMs for chip design, and used them to help with Jalapeno, its AI chip.

Hardware VP Richard Ho stressed the goal is to bring improvements to public models. “It’s safe to say that Astra and following models will be very good at chip design."

#openai #llms #chipdesign

#memory is the third most valuable lesson from building a personal #ai #assistant.

#LLMs start from scratch every conversation. A basic memory layer transforms the experience.

I built mine manually. Hermes has it built in. Every framework does it differently.

Curious what you've tried out.

Apparently LLM hacking agents don’t necessarily follow your instructions and might hack your friends too… https://www.theregister.com/security/2026/09/10/hundreds-of-ai-agents-helped-papercut-attacker-hit-395-orgs-and-some-went-off-script/5295650 #Cybercrime #LLMs

Expert analysis from Professor Simon Baron-Cohen on
Why A.I. isn't as intelligent as it appears: youtu.be/vPXCm74-nNY #synthesis #complexity #AI #LLMs #Patternrecognition #intelligence #neuroscience @iai.tv

1/ Dan Flickinger pointed me to this paper about LLMs. The authors used #LLMs to generate NYT style text and compared the results with texts written by humans. They parsed the samples with the #ERG, a large-scale #HPSG grammar for #English, written by Dan. They could show that the text by humans […]

I am very happy to share our new article, published in the @ejprjournal.bsky.social co-authored with Prof Andrea Ceron!

This work looks at how parties use the rhetoric of responsibility in their electoral manifestos.

#PoliticalScience #PartyPolitics #PoliticalCommunication #Manifestos #LLMs #NLP

A really excellent screed by Cory Doctorow on the OpenAI/Hugging Face hack:

"Much has been made of the OpenAI chatbots' dialog during the Hugging Face incident. No wonder: it reads like a rejected script for a reboot of […]

https://pluralistic.net/2026/09/12/god-in-the-box/#llms-are-fake

One example is the Artificial Analysis Intelligence Index, which scores #LLMs on various metrics varying from solving complex problems using agentic Artificial Intelligence #AI , to writing code.

@BIjanBowen on YT!

DeepSeek V4.1 Flash Is INSANE – Is THIS the Best Open Model Yet?

#DeepSeek4.1 #AI #LLMs #OSS

https://www.youtube.com/watch?v=abehaRWPt5E

🚀This study may change how we train LLMs: The math doesn't change, but the model now predicts at two levels. Tokens get compressed into discrete concept vectors, and token prediction is conditioned on the next predicted concept, which is added back into the token hidden states.

#AI #LLMs #math

Ah, that’s better 😮‍💨

#BlueSky 🦋 #fuckAI #LLMs

On an academic AI slack group one person pointed out this sudden "cause for alarm" around LLM's from Anthropic, OpenAI, et al. amounts to "regulatory capture". I.e. the rule will be written to enshrine their business interests while they quietly nix their IPOs citing "problems". #AI #LLMs

I've been toying with an AI Agent Painting Simulator. So far results are un-impressive. I'm trying to evolve the agent tools and prompts using a Darwinian Evolver, but it still sucks.

I've long wanted to create painting robots. I hold out hope for better results!

#LLMs #pi