msp26

9 ms·

91.
by msp26 2y ago
> one of our reasons for launching products early and often is to give society and the technology time to co-evolve Is this really true? O3 (not mini) is still being held for ""safety testing"", and Sora was announced
92.
by msp26 2y ago
How well do they work when you want to do things like grounding with search?
93.
by msp26 2y ago
> a much more helpful and detailed version of this Notice the deliberate wording. To me this implies we aren't getting the raw CoT.
94.
by msp26 2y ago
I wish they'd just reveal the CoT (like gemini and deepseek do), it's very helpful to see when the model gets misled by something in your prompt. Paying for tokens you aren't even allowed to see is peak OpenAI.
95.
by msp26 2y ago
Finally, all the recent MoE model releases make me depressed with my mere 24GB VRAM. > Note that Mistral Small 3 is neither trained with RL nor synthetic data Not using synthetic data at all is a little strange
96.
by msp26 2y ago
How can openai justify their $200/mo subscriptions if a model like this exists at an incredibly low price point? Operator? I've been impressed in my brief personal testing and the model ranks very highly across most benchmarks (wh
97.
by msp26 2y ago
The cost of LLM inference is cheap and will continue to decrease. More traditional methods take up far more of an engineer's time (which also costs money). If I have a project with a low enough lifetime inputs I'm not wasting my t
98.
by msp26 2y ago
Definitely buy a pull up bar, it's one of the best purchases I've made. Pull ups and chin ups are fun exercises and they make your back feel REALLY good.
99.
by msp26 2y ago
> copy You mean build on existing public research? Everyone does that. At least deepseek, meta etc. also have the decency to publish research back into this ecosystem.
100.
by msp26 2y ago
Western LLM censorship affects me far more than Chinese LLM censorship.
101.
by msp26 2y ago
Actually having read it now, there's way too much fluff. This could be way shorter.
102.
by msp26 2y ago
Looks fantastic, thanks for the deep dive on structured output. Will read thoroughly.
103.
by msp26 2y ago
It depends on where you're looking haha.
104.
by msp26 2y ago
I agree with you completely. I was talking about the parsing being easy with this, not referring to the outputs being correct in reality. You can get awful results with poorly defined constraints.
105.
by msp26 2y ago
It really isn't necessary when using constrained decoding (aka structured outputs) which guarantees that you'll get JSON output in the correct structure.
106.
by msp26 2y ago
> which I suspect harms accuracy over free form Untrue in my testing. If you want to use chain of thought, you can always throw in a `thoughts` field (json field/xml tags) before the rest of your output.
107.
by msp26 2y ago
I've had a situation where Claude (Sonnet 3.5) refused to translate song lyrics because of safety/copyright bullshit. It worked in a new chat where I mentioned that it was a pre 1900s poem.
108.
by msp26 2y ago
It's incredibly well written. I can see this being very helpful for newcomers. As for the Predicted Outputs feature, it looks incredibly useful in a few of my pipelines. Can't wait to test it out.
109.
by msp26 2y ago
But doesn't this fully lock you into using OpenAI's offerings? If you store the finetuning dataset yourself, you are free to use it on whatever model/provider/self hosting. And I'm finding it fairly straightforward
110.
by msp26 2y ago
Just out of curiosity, what sort of challenges did you run into when scaling this up? I don't see a need for my current solution to go past a handful of browser instances but I'd imagine it might get crazy.
111.
by msp26 2y ago
Awesome, I've been working on a similar thing at a smaller scale and I think this area is very promising. I've limited my problem scope to single page interactions / scraping which has been very reliable and useful for my com
112.
by msp26 2y ago
Ridiculous blocking
113.
by msp26 2y ago
Have you looked into structured generation with a library like outlines? https://github.com/dottxt-ai/outlines
114.
by msp26 2y ago
Any chance of this being released open weights? Or is the risk of bad PR too high (especially near a US election)? It being 30B gives me hope.
115.
by msp26 2y ago
Has anyone tried Google's context caching feature? The minimum caching window being 32k tokens seems crazy to me.
116.
by msp26 2y ago
I'm surprised to hear that. I've genuinely never held so much disdain for a TV show (season 1) in my life. But when season 2 hit I was too apathetic to even bother watching. They made so many changes (mostly shit) that it was hard
117.
by msp26 2y ago
It's a breath of fresh air watching anime adaptations that actually love and respect the original author's work like dungeon meshi. Meanwhile western fantasy adaptations seem to be full of arrogant showrunners that think their vis
118.
by msp26 2y ago
> If your knowledge base is smaller than 200,000 tokens (about 500 pages of material) I would prefer that anthropic just release their tokeniser so we don't have to make guesses.
119.
by msp26 2y ago
What would you recommend for parsing instead?
120.
by msp26 2y ago
Shoutout to Elden Ring where the entire game freezes when your (wireless) mouse goes to sleep.

More ›