zaptrem

11 ms·

121.
by zaptrem 2y ago
Yes, but if you want more usage it is reasonable to expect to pay more.
122.
by zaptrem 2y ago
I have substantial usage via their API using LibreChat and have never run into rate limits. Why not just use that?
123.
by zaptrem 2y ago
Yes it’s important to make sure it’s easy to verify the code is correct.
124.
by zaptrem 2y ago
I use these models to aid bleeding edge ml research every day. Sonnet can make huge changes and bug fixes to my code (that does stuff nobody else has tried in this way before) whereas GPT 3.5 Turbo couldn’t even repeat a given code block wi
125.
by zaptrem 2y ago
I meant "what incentive would they have to just now start discriminating against their high performers?" This is a multi trillion $ mega-corp with thousands of metric and profit-focused decision makers, not Judd's Confederate
126.
by zaptrem 2y ago
What incentive would they have to do that?
127.
by zaptrem 2y ago
In autoregressive models error accumulates over time. He likely means the voice starts to make odd sounds/gets lower quality. It would be really interesting if OP could share a clip of this phenomenon!
128.
by zaptrem 2y ago
Decentralize iCloud doesn’t solve the problem because the majority of users (myself included) would want to keep using their hosted/“official” version due to the Apple subscription bundles we’re already paying for (the blog mentions th
129.
by zaptrem 2y ago
"You can train a SOTA LLM for $0.50" (as long as you're distilling a model that cost $500m into another pretrained model that cost $5m)
130.
by zaptrem 2y ago
This is how I think about unlimited data plans haha. I think a rate limit is easier to stomach (e.g., X requests per hour where they bank up to one hour so you can burst up to 2X for especially crazy hours or something).
131.
by zaptrem 2y ago
Transformers are deep feedforward networks that happen to also have attention. Causal LMs are super memory bound during inference due to kv caching as all of those linear layers need to be loaded onto the core to transform only a single t
132.
by zaptrem 2y ago
The models are not self aware of their training data. They are only aware of what the internet has said about previous models’ training data.
133.
by zaptrem 2y ago
Problem is microplastics
134.
by zaptrem 2y ago
Haven't we had this for months now? I've used all of these already (though maybe they're just in the developer betas?)
135.
by zaptrem 2y ago
It is a lot easier to switch LLMs than it is to switch smartphone platforms.
136.
by zaptrem 2y ago
Have there been compelling use cases for on device LLMs where the value isn’t based on privacy benefits yet?
137.
by zaptrem 2y ago
Assuming it was sold profitably wouldn't this just cause more to be built?
138.
by zaptrem 2y ago
NYC's defining trait and constant throughout history is development. The concept of a tiny row houses and a car on every driveway in Queens was invented in the mid 20th century when Robert Moses scarred a half dozen neighborhoods to ma
139.
by zaptrem 2y ago
We wondered the same thing! I wonder which model they’re using.
140.
by zaptrem 2y ago
We used it to make memes (single sentence prompts) and the results are hilarious How Jeb! Won the election: https://math-gpt.org/?video_id=b9e16dbd-cab6-44e3-a393-f255b... Mathematical proof AI is doomed: https:/&#x2F
141.
by zaptrem 2y ago
Idk what incentive they would have to do that given iirc they’re selling basically every EUV machine they can make right now?
142.
by zaptrem 2y ago
O1, new Sonnet, all the music models and video models, the voice models like 4o, etc.
143.
by zaptrem 2y ago
This would have been interesting if true, but this other HN comment suggests it isn’t https://news.ycombinator.com/item?id=9374977
144.
by zaptrem 2y ago
Isn't Trump the pro-"crypto Ponzi scheme" (as you called it) candidate? Asset prices of both crypto and stocks seem to think so.
145.
by zaptrem 2y ago
Why did they mention "gaming performance" (i.e., frame rate change which isn't always directly connected to CPU performance) and not absolute performance (across a range of benchmarks)? The latter is more meaningful.
146.
by zaptrem 2y ago
IMO ray tracing remains a gimmick and the die space would be better spent on more normal compute hardware. Basically every game that doesn't have a day/night cycle can prebake their lightning and have it look nicer than any possib
147.
by zaptrem 2y ago
This is almost completely undocumented outside of this HN post (check Google). Can you tell us more about what this does and how it is intended to be used? Thanks!
148.
by zaptrem 2y ago
While I originally understood it the way you did, I think they were referring to this: https://en.wikipedia.org/wiki/Emancipation_Proclamation
149.
by zaptrem 2y ago
Could OpenAI make it impossible for startups to try to build AI search engines by signing all these paid agreements with publishers?
150.
by zaptrem 2y ago
Neat! Are these compatible with torch compile?

More ›