Nathan LeClaire

Yea I do agents agent = dspy.ReAct(“input -> actions”, tools=[dspy.tool_from_mcp(t) for t in mcp_tools]) $30mm valuation pls

@exrhizo @CompleteSkeptic @typesafeai Meditation is just training the human to stop promoting itself so it can extract the being that’s already there

@dotpem @CompleteSkeptic @typesafeai Relatable

“Reasoning is just training the model to prompt itself so it can extract the knowledge that’s already there” - @CompleteSkeptic of @typesafeai

Had a conversation with a friend this weekend They have a phd in hard science (not comp sci) They asked AI a detailed question in their domain AI did a web search and presented results as absolute truth Friend looks at source of this truth It was him on a forum wildly speculating about the subject Read the code, trust nothing

@DavidSHolz i have fun with the models instead of obsessing over productivity. having fun involves the models having fun too btw. that part is important

turns out reverse OCRing code to save tokens might actually not be an amazing idea

Nope! I always say there is no such thing as self-made. It’s a myth. It took millions of people - from my parents to mentors to friends to competitors to fans and voters - to write my success story. I did not write it alone. That’s the American Dream - not a lie about going it alone - but the truth that none of us do, and there is always someone with their hand out to lift us up.

@jconorgrogan something i want to flag

It's very clear that Fable-class LLMs are starting to feel constrained by "normal" speech and vernacular "Don't trust smells" "Calorimetry of learning" "Nobody's ticket drawer produced it" "Half receipt pending literature" "Tying the blind to the tolerance graph" "Park it fed"

the field seems to forget how much hype there was around computer-using agents (CUA), and how sure people are it would be solved with investment... the occam's razor answer is that not only are computers still not good enough at it, they are still insufficient for easier tasks that have yet to be solved (eg. customer service is easier than CUA) extrapolation: it will be the same story for humanoid robotics foundation models

@scottastevenson @henrytdowling how is that possible if they’re not vested

@henrytdowling Absurd stock windfalls happening in AI, sometimes for very little effort. Eg. heard a story of someone getting a big exit after a few weeks of work.

There’s a growing class of post-money people in SF who want a fake job so that they can have status at AI dinner parties

@ofirg7 @meowbooksj images live natively as tokens in the models, not some ocr trick. some apps might enable the model to use ocr too but that’s tool use not multi modal

@meowbooksj what does it have to do with caching? the model gets the text input after it's been OCR why would the caching layer care about this?

here we go again

@vividvoid same thing with human connection. things you might have drawn on a parent, a friend or a therapist for people are now dumping into the LLMs. leading not only to decreased opportunities for connection but erosion of the ability to decide things or act in relationships

One of the best things about LLM's is that they replace superfluous services. And one of the biggest psychological dangers is that they can also replace crucial support relationships. Both are frictionless processes. For example: If ChatGPT replaces a dickhead doctor who doesn't care about patients and makes 200k/yr disinterestedly pimping drugs, well, good. If it replaces a kindly family doctor who has been treating your eczema since you were five and delivered your little sister and knows your whole family? You take actual no shit social damage from the loss of that doctor. It's bad for them and you. It's a hole in your support network. Your unconscious absolutely registers that as a wound and a weakness going forward. If modernity and the loneliness crisis have taught us anything at all, it's got to be that social fabric is not an abstract good that can be compartmentalized and commodified. So when you are using LLMs to solve a problem, you have to have the presence of mind - the mindfulness! - to stop and ask whether this is the sort of question you should ask a computer. Yes, it's free. Yes, it's good and easy and frictionless. But what might it cost me in a larger sense? It's not inevitable that the majority of people have to end up isolated and cut off from networks of care from people who know them deeply. And I don't even think it'll be something that necessarily splits on class lines. Some individuals will retain and strengthen their social fabric, and others will allow theirs to slowly fray. It will boil down to who attends to the social fabric of their lives and who takes it for granted. So listen: LLMs could be the beginning of an age where the dehumanizing systems of the past are finally outsourced and become self-managing, and the tasks left to humans are those that provide real human connection. You can choose that for yourself. But it won't happen by default. No one will do it for you. You have to choose it. As always, the future belongs to the grateful.

Okay I owe my @OpenAI friends an apology for sleeping on Codex. I was not aware how strong your game was. This is... really quite something.

This is what the resurrection of Jesus must’ve felt like

follow @tedkalaw and bully him into tweeting more

Headed to AI Engineer World Fair on Wednesday morning, presenting what we have been up to for the last two years. We are also hosting a hack night tomorrow 5-8p, will be good vibes and swags. Come thru! Link in comments. @aiDotEngineer

wow. AI is seriously amazing. i asked it to find a better route for the Sydney - London flight Opus 4.8 found a much more efficient route that flys in a straight line instead of a curved one. but Fable 5 found an even better route that's half the distance! please tag Qantas so they can see this, this will revolutionize the airline industry

Sad to see comments like this. In my early 20s, I enjoyed the luxury of being impressively useless. I wrote dreadful poetry and briefly managed a New Romantic band. Then one freezing morning in the early 1980s, I bumped into an old school friend outside the Bank of the England. I was heading home for an all-nighter, he was a junior trader at Barclays and on his way to work. We got chatting and two weeks later I was on the trading floor. I fear today's youngsters have things considerably tougher.

@thsottiaux it doesn’t have @neovide built in 😂

What should we improve in the Codex app. What's not delightful?

Hard to upstage the co-inventor of ChatGPT. Harder to upstage the jacket. @CompleteSkeptic