Earlier quoted context omitted.
not your weights, not your bitcoins
now its 18. iykyk
care to explain?
GPT-4 API General Availability
341–350 of 562 posts
Re: GPT-4 API General Availability
#342>Developers wishing to continue using their fine-tuned models beyond January 4, 2024 will need to fine-tune replacements atop the new base GPT-3 models (ada-002, babbage-002, curie-002, davinci-002), or newer models (gpt-3.5-turbo, gpt-4). Once this feature is available later this year, we will give priority access to GPT-3.5 Turbo and GPT-4 fine-tuning to users who previously fine-tuned older models. We acknowledge…
This tells me that either there were very few commercial users of finetuned models, or they need to decommission the infrastructure to free up GPU's for more valuable projects.
There’s also the possibility that they weren’t seeing lots of ongoing usage of existing fine tuned models e.g. users tuning, running some batch of inputs, then abandoning the fine tuned weights.
Re: GPT-4 API General Availability
#343Earlier quoted context omitted.
A big enough hole presents a wedge for new entrants to get started. OpenAI will never fulfill the entire market, and their moat is in danger with every other company that has LLM cash flow. They want to become the AWS of AI, but it's becoming clear they'll lose generative multimedia. They may see the LLM space become a race to the bottom as well.
Let's hope so - the amount of control they have over this is a great evil. Many of us have experienced the potential of a less moderated GPT4, and we all know that somewhere out there, they have the full unmoderated version. What are they using it for? What powers have got their hands on this thing?
That's exactly what I'm thinking right now.
Re: GPT-4 API General Availability
#344Earlier quoted context omitted.
The llm command-line tool looks great: https://llm.datasette.io/en/stable/
are we out here typing our api keys into random pips and am i a boomer that i would be hesitant to do it
It’s not a “random pip”. The maintainer is a well-known open source developer (one of the creators of Django and Datasette). It’s also a very small codebase – not many places for malicious code to hide.
Re: GPT-4 API General Availability
#345Earlier quoted context omitted.
If it really was a tiny number of users, they would publically make a really good offer - for example: "Unfortunately, you will need to retune your models on top of GPT-4. OpenAI will do this for you for free, and refund all money you paid tuning your original model, and offer the new model for the same price as the original model." The extra trust gained by seeing another customer treated that way easily pays for a…
OpenAI probably doesn't feel the need to pay to win publicity right now—they've been in the spotlight for as long as LLMs have been a thing, and GPT-4 is far ahead of competitors' offerings.
It’s about trust - not publicity. Trust is hard to earn back once broken, and there will be multiple offerings eventually.
For example, AWS was one of the first cloud providers. Now there are alternatives, but I still pick AWS because I trust them not to break my dependencies way more than, say, Google
Re: GPT-4 API General Availability
#346Does this mean we don't have to give OpenAI our phone number anymore?
Re: GPT-4 API General Availability
#347The drip-feeding seems crazy to me. Open AI is undermining their reputation by forcing almost everybody to use the older, lower-quality models. Even if customers are willing to pay for GPT 4, they're being told to wait at the back of the line. Wait for what!? Christmas? When we can open our presents and have a GPT 4 inside? It's like they took a leaf from Google's "how to guarantee the failure of a new product" marke…
I think maybe you don't understand that they don't have enough GPUs to do this, and money can't buy enough GPUs to do it.
This is the bottleneck. EUV Photolithography is one of the hardest engineering challenges ever faced, it's like trying to drop a feather from space and guaranteeing it lands on a specific blade of grass. Manufacturing these GPUs at all requires us to stretch the limit of what is physically possible in multiple domains, much less producing them at scale.
Re: GPT-4 API General Availability
#348Biggest news here from a capabilities POV is actually the gpt-3.5-turbo-instruct model. gpt-3.5-turbo is the model behind ChatGPT. It's chat-fine-tuned which makes it very hard to use for use-cases where you really just want it to obey/complete without any "chatty" verbiage. The "davinci-003" model was the last instruction tuned model, but is 10x more expensive than gpt-3.5-turbo, so it makes economical sense to hack…
I'm interested in the cost of gpt-3.5-turbo-instruct. I've got a basic website using text-davinci-003 that I would like to launch but can't because text-davinci-003 is too expensive. I've tried using just gpt-3.5-turbo but it won't work because I'm expecting a formatted JSON to be returned and I can just never get consistency.
Have you tried guidance?
Re: GPT-4 API General Availability
#349Earlier quoted context omitted.
I think maybe you don't understand that they don't have enough GPUs to do this, and money can't buy enough GPUs to do it.
This is the bottleneck. EUV Photolithography is one of the hardest engineering challenges ever faced, it's like trying to drop a feather from space and guaranteeing it lands on a specific blade of grass. Manufacturing these GPUs at all requires us to stretch the limit of what is physically possible in multiple domains, much less producing them at scale.
Thanks for this explanation! :) (as someone without knowledge of the hardware process I appreciated it).
It is SO amazing that we have such a driving force (LLMs/consumer-AI) for this (instead of stupid cryptocurrencies mining or high-performance gaming). This should drive innovation pretty strongly and I am sure the next "leap" in this regard (processing hardware) will put technology in a completely different level.
Re: GPT-4 API General Availability
#350The difference between 4 and 3.5 is really big for creative use cases. I am running an app with significant traffic and the retention of users on GPT-4 is much higher. Unfortunately it's still too expensive and the completion speed is not as high as GPT-3.5 but I hope both problems will improve over time.
You might be able to make it into a ChatGPT Plugin and then you don't have to pay for that part of the completion.