Kenan

@hamptonism Just because it follows outer product arithmetic doesn’t not mean it’s matrix theory.

The origins of linear algebra can be traced to ancient China, where in 240 BCE, the philosopher Xian Wáng developed an early form of matrix theory to describe the interactions between the Five Elements-Metal, Wood, Water, Fire, and Earth.

@OpenAINewsroom Will this information be available via API calls to 4o?

We’re partnering with Hearst to bring their iconic curated lifestyle and local news brands to OpenAI’s products. openai.com/index/hearst/

@tsarnick Models are of fixed size. Every time you pass an input in test-phase, the entire network activates, and it’s always the entire network. Hence, amount of compute is always the same. How can training for longer reduce compute?

OpenAI's Noam Brown says that while AI model performance scales roughly equivalently with more training or inference compute, the cost of inference is on the order of 100 billion times cheaper

@michael_kove @flowersslop Per OpenAI API, the model does not have access to metadata.

@flowersslop Show me the metadata you sent with that photo? You took it on your phone? Ask ChatGPT if it has meta data parsed from the photo.

ChatGPT can identify the exact Caribbean island from a photo of a parking lot That seems impossible - I mean, look at that photo. That could be fucking anywhere in the world. And it doesn't just narrow it down to the Caribbean islands, it also names the EXACT Caribbean island

@paulg @ATabarrok No, because it’s hard to distinguish if you got a negative result or simply fucked up.

@ATabarrok When existing journals ignore some class of discoveries, isn't the solution to create new journals? Related but not identical idea: would it be a good idea to start a journal specifically focused on negative results?

@jeremyphoward According to FAIR, avoiding markdown improves LLM performance.

Increasingly, docs are for LLMs to read, not just for humans to read. We're starting to explicitly design our docs to be great for *both* uses. As part of that, I'm aiming to include full markdown docs in our projects. FastHTML is the 1st! Here you go: github.com/AnswerDotAI/fa…

@biogerontology Is 380 high or low for this field?

Can you imagine that the article called "Total synthesis of rapamycin" published in 1991 has only around 380 citations. All synthetic papers that came after - even lower

@fchollet Lol

You can tell a *lot* about a man by watching him interact with his wife/gf/partner for 20 seconds. I cannot think of a more information-rich signal.

@svpino Its suggestions, when wrong, are distracting.

If you aren't using GitHub Copilot to write code, can you tell me why?

@radshaan @PandoXiloscient Formal Logic is the way. Phil005 at UPenn.

@PandoXiloscient I want more structure in my thought process, that is my goal

I have never read philosophy, what’s a good way to start?

@williamwei @drfeifei @StanfordHAI @TIME You are right.

@drfeifei @StanfordHAI @TIME I believe we will pass the super-intelligent stage before we achieve sentience.

Is AI sentient? My friend and colleague Prof. John Etchemendy, a renowned professor and co-Director of @StanfordHAI , just co-authored this piece to debunk the claim that today’s LLMs are sentient @TIME time.com/collection/tim…

@realGeorgeHotz The difference is that humanoids will be used in factories which can be mapped accurately. Real-world roads cannot.

@escliu Lol

ok it has reached that point - im looking for a stylist in SF. recs?

@evitable_cnflct @paulg True. Didn’t realize that.

@paulg I’m long on AI as general interface. Hard to see a world in 5 yrs where building CRUD apps is necessary. Apps will remain for specialised domains, but browsers are going the way of gopher. Timelines are hard to judge, but how many browser tabs are already MIA from ChatGPT et al.

This may seem like a type error, but I wonder if AI is the next form factor in computing.

@petewilz @paulg Thats exactly it.

@paulg I think it’s going to be a shift from Personal Computers to Enterprise Computers. Apps were the atomic unit in PCs; I think agents will be the equivalent in the enterprise OS. I think we will look back and think of the LLM as a new interface like the GUI.

@spirit_raccoon @paulg @matthew_d_green Why is efficiency the matter? We have plenty of energy.

@paulg @matthew_d_green just shows how fixed everyone is on the current thing...openAI is great at staying a few months ahead but the winner will be whoever can figure out how to set up an LLM as efficiently as biological systems.

Why do people keep talking about OpenAI as though they have some unique secret sauce, when basically every AI company with a GPU budget is able to replicate their achievements within a few months?

@auren What is go-forward comp?

Founder CEOs: your go-forward compensation should be at least 70% of what an outside CEO would make. (regardless of your already-vested ownership) if you are getting less, you are getting screwed by your VCs

@SullyOmarr @NickADobos @vercel Share the script please.

@NickADobos @vercel Free for now and I just made a script that did it

Gemini 1.5 pro is STILL under hyped I uploaded an entire codebase directly from github, AND all of the issues (@vercel ai sdk,) Not only was it able to understand the entire codebase, it identified the most urgent issue, and IMPLEMENTED a fix. This changes everything

@daltonc ^big problem in fashion retail. A solution (1-min): youtube.com/watch?v=4W95S6…

Overestimating demand for a product sure seems to be a cause of death for a lot of companies. Likely because the twin problems of not having enough supply to satisfy demand vs overestimating demand and overbuilding are not equally easy to fix.

@paulg Granted the peoples of that country have the capacity to organize, which is hard.

When I was a kid I thought Naziism was connected with the German character. Nazis in WW II movies were always so German. But the more I see, the less plausible this seems. The people of any country can behave like Nazis if they get into a bad situation with bad leaders.