minimaxir

7 ms·

151.
by minimaxir 3mo ago
I did not flag the original HN submission, but yes, this post reads very weirdly at best to the point that it's hard to take it seriously. Not every paragraph needs to be 2 sentences that are snippy.
152.
by minimaxir 3mo ago
You likely got downvoted/flagged here ( https://news.ycombinator.com/item?id=49140552 ) because it is disproportionately aggro and against HN guidelines. Also you can't do the "this used to be an actual place w
153.
by minimaxir 3mo ago
It doesn't add anything to the discourse. If it's low-quality, flag it and move on. (the fact that it's currently at #3 is a bit weird and I've noticed some things weird recently with the HN algorithm but that's off
154.
by minimaxir 3mo ago
Interesting that this post is credited to Sarah Friar, OpenAI's CFO, and not a CEO/CTO/CMO.
155.
by minimaxir 3mo ago
Most of the startups mentioned are YC companies.
156.
by minimaxir 3mo ago
I do not casually have spare tactile switches, nor the hardware skills to replace.
157.
by minimaxir 3mo ago
My G602 died due to Logitech's infamous double-click issue. :( I've been fine with the G502 X Lightspeed (which should avoid said double-click issue due to optical switches) and some more efficient button use to compensate for t
158.
by minimaxir 3mo ago
...does Hugging Face have enough bandwidth to let people download en masse however much file size a 2 trillion parameters model is?
159.
by minimaxir 3mo ago
The obvious solution is to just send half an email!
160.
by minimaxir 3mo ago
I once had an interview loop where after the final round I didn't hear back from the recruiter point-of-contact for awhile. After 1.5 weeks, I sent an email following up: the recruiter replied "I was on vacation dammit, btw you di
161.
by minimaxir 3mo ago
> It's not very clear why using GPT-4 in early 2023 was so expensive and then six months later 20 bucks could get you a fair amount of GPT-4 inference. FlashAttention was a hell of a drug.
162.
by minimaxir 3mo ago
Those transcripts generally contain code edit diffs.
163.
by minimaxir 3mo ago
A notable surprise with Bluesky over the past few months is that the anti-AI sentiment has decreased slightly as it becomes harder to argue against impactful use cases.
164.
by minimaxir 3mo ago
"I also had a good experience providing my full genetic sequence" is a very funny sentence out of context.
165.
by minimaxir 3mo ago
This was announced in January ( https://openai.com/index/introducing-chatgpt-health/ ) but this now appears to be the GA rollout.
166.
by minimaxir 3mo ago
Landing pages were already heavily templated prior to agents.
167.
LLMs break down in funny ways when told the Jacobian Conjecture counterargument(minimaxir.com)
1 points by minimaxir 3mo ago|0 comments
168.
by minimaxir 3mo ago
Relatedly, Codeberg has also banned vibe-coded projects. https://news.ycombinator.com/item?id=49003386 https://news.ycombinator.com/item?id=49003465
169.
by minimaxir 3mo ago
I initially had a Rust-based word cloud generator that generates word clouds in high resolution in ~100 milliseconds whereas it would take other generators a couple seconds to do the same thing. Does the world need a super-fast word cloud g
170.
by minimaxir 3mo ago
Both the example libraries compared (tokenizers and tiktoken) are Rust-based with Python bindings. There's just a few levers in Rust that can speed it up even more particularly with LLM assistance as the AI Use Discloure here notes:
171.
by minimaxir 3mo ago
> just repeatedly saying "keep going" to ChatGPT For posterity, this indeed works for most problems where an agent might give up. LLMs don't inherently know something is impossible. The phrase I tend to use in my harder
172.
by minimaxir 3mo ago
I'll have a blog post up tomorrow about it but the Jacobian Conjecture counterexample is a very funny cognitohazard for LLM assistants. It's a paradox for modern LLMs: they have enough math skills such that they can easily compute
173.
by minimaxir 3mo ago
It's a meme. Telling it to "make no mistakes" doesn't do anything because LLMs don't have an inherent concept of a mistake and they are already RLHFed to code correctly. However , if you tell it to not do particula
174.
by minimaxir 3mo ago
Also note the timestamp: he started working on this thread a few hours after the tweet.
175.
by minimaxir 3mo ago
This proof was posted by an individual researcher, Anthropic has not used it in any of their marketing. Additionally, posting this during the World Cup would be the most inefficient way to do marketing.
176.
by minimaxir 3mo ago
There will be more and more mathematical proofs provided by LLMs as they improve. If you assume every one is just a marketing tactic, you'll drive yourself insane.
177.
by minimaxir 3mo ago
> I think the argument is sound They got flagged because it is a literal conspiracy theory that assumes bad faith.
178.
by minimaxir 3mo ago
> So the conjecture that survived Keller, Abhyankar, Moh's degree-100 verification, and five-plus published wrong proofs appears to have died via tweet during the World Cup final.
179.
by minimaxir 3mo ago
Who knew DOES NOT COMPUTE would be an actual thing?
180.
by minimaxir 3mo ago
You accused me of "lies I'm telling myself" about the robustness of current LLMs and I take offense to that when I have hard evidence that it's not a lie I'm being manipulated by. That's not deflection and I&#x

More ›