vatsachak

8 ms·

211.
by vatsachak 3mo ago
But each company itself is composed of multiple agents. It's just that companies are a scale of consolidation. The exact same thing happens in the cortex, 10s of thousands of cortical columns coalesce into a single action.
212.
by vatsachak 3mo ago
GPT 5.6 Sol is the first model I've used where I can trust it to add 100-500 lines of code maintainably. It's great with Codex. I still find that LLMs tend to not know how to compose larger ideas but on the scale of small ideas or
213.
by vatsachak 3mo ago
How could they? That's why the free market works, millions of agents in parallel beats any planned economy (by humans)
214.
by vatsachak 3mo ago
You can embed TLA+ into Lean. I don't think that there's any benefit to formalization before coding unless you're working on like million dollar projects
215.
by vatsachak 3mo ago
I don't really buy this argument because we can all read the code with the Lean LSP. Also, after using a tactic enough you can guess why it's used. Agda and Idris are more beautiful for sure, but a proof is a proof (according to t
216.
by vatsachak 3mo ago
Terrence Tao's blog is better than most textbooks and everyone has been using it for the past 20 ish years lol
217.
by vatsachak 3mo ago
Yeah more or less. I proved a SOTA result using Gemini 3.1 Pro a year ago and it was a lot of back and forth. We're definitely still in the computer chess phase.
218.
by vatsachak 3mo ago
And yet it can't even replace a call center employee. I guess you can't make a career out of being a frog anymore in math.
219.
by vatsachak 3mo ago
Yeah. But we need throw money at studying the brain, instead of another billion to "agents which understand you truly"
220.
by vatsachak 3mo ago
Super big fan of your work on cellular automata. Do you have anything interesting papers coming up?
221.
by vatsachak 3mo ago
Interesting. But, evolution is way too unconstrained to provide us a path to "agi". It would require too much compute. Evolution also eventually gets frustrated and creates the brain, capable of in context learning. Maybe we shoul
222.
by vatsachak 3mo ago
Maybe a subset of mathematicians, but if someone proved that RH was undecidable we would still give them the millennium prize.
223.
by vatsachak 3mo ago
Very fair. But when you say "What mathematicians care about", you are taking about mathematicians today, who really care mostly about politics
224.
by vatsachak 3mo ago
I think that your take is quite optimistic. Having published in top tier journals my only experience is that mathematicians care about what other mathematicians worked on and failed to solve. Theory building papers are dime a dozen and don&
225.
by vatsachak 3mo ago
It's not a problem of incentives. Every executive wants to inject LLMs everywhere these days. If they haven't somewhere it means that it does not work.
226.
by vatsachak 3mo ago
Not always, sure but 90% of the time yes. For example, create a DFA for a regex, not too bad just use Thompson's algorithm and then NFA->DFA. But now we have to care about efficiency, user API, maintainability of definitions etc. Co
227.
by vatsachak 3mo ago
No it can't. Show me a business which uses in context learning to manage a McDonald's
228.
by vatsachak 3mo ago
Math is way more automatable than programming. In math, a proof is a proof. We don't know if we can get there and so getting there is the hard part. In software, we always know that we can solve the problem. So HOW to solve the probl
229.
by vatsachak 3mo ago
Yeah pglite is exquisite. Now I don't have to write separate code for client and server side queries. For example, I have a compiler that compiles a DSL to a DB query which returns a list. Now that query can either be on data in the br
230.
by vatsachak 3mo ago
Literally just keymaps
231.
by vatsachak 3mo ago
Here's my two cents as a mere paper reader; JEPA is really just a generalized encoder, so the JEPA created latents should be fed into a transformer trained with either user data or RL policy. Although the above might not work great eit
232.
by vatsachak 3mo ago
Automated coffee vending machines are not at the level of barista. Plus a coffee shop is about the experience, not the price. But most jobs that aren't service are not about the experience.
233.
by vatsachak 3mo ago
The world has been getting richer as everything has gotten automated
234.
by vatsachak 3mo ago
RSI isn't anything new though; computers have been used to make computers better for about 80 years now. Imagine having a secretary who could read 1 million records and give you back your answer in 100 microseconds, for just 10 cents a
235.
by vatsachak 3mo ago
Now that's what we call bait
236.
by vatsachak 3mo ago
I don't hate SF it's just overpriced. Whenever I visit SFO it's really funny seeing all the advertisements from startups above a population struggling to find housing. Won't it be better to pay someone 100k in Reno than
237.
by vatsachak 3mo ago
Creativity can be automated. Humans are automated creativity invented by evolution
238.
by vatsachak 3mo ago
I find that Amazon is still the best for new, niche items like stationery. Maybe I need a Bayesian update
239.
by vatsachak 3mo ago
Yeah. Target has been more than pleasant, but I know that they have always been good at design and ergonomics. I should probably just switch to them
240.
by vatsachak 3mo ago
Using LLMs to generate dashboards is probably their most productive use case

More ›