Comment by gzer0 | Hacker News Reader

Comment by gzer0 | Hacker News Reader

- Context length extended to 128k (~300 pages).

- Better memory retrieveal across a longer span of time

- 4 new APIs: DALLE-3, GPT-4-vision, TTS (speech synthesis), and Whisper V3 (speech recognition).

- GPT-4 Turbo, a more intelligent iteration, confirmed as superior to GPT-4.

- GPT-4 Turbo pricing significantly reduced, about 3 times less expensive than GPT-4. Input and output tokens are respectively 3× and 2× less expensive than GPT-4. It’s available now to all developers in preview.

- Improved JSON handling (via JSON mode) and function invocation for more sophisticated control.

- Doubled rate limits with the option to request increases in account settings.

- Built-in retrieval-augmented generation (RAG) and knowledge current as of April 2023.

- Whisper V3 to be open-sourced and added to the API suite.

- Copyright Shield initiative to cover legal fees for copyright-related issues.

- Ability to create your own, custom "GPTs".

- Assistants API and new tools (Retrieval, Code Interpreter)

The Whisper V3 model was released open source a few minutes ago: https://github.com/openai/whisper/commit/c5d42560760a05584c1...

Nice. So it's just a new checkpoint for the large model, no other changes? Should be easy to integrate, though it would be nice to have updates for the smaller models too.

This is much better than I would have thought, especially pricing. It may also be a sign that OpenAI expects competition in the near future, which would be another good news.

I just had a hope that the GPT 4 API was a little bit more cheaper

> It may also be a sign that OpenAI expects competition in the near future, which would be another good news.

Of course there is competition. OpenAI is using the Andrew Carnegie playbook of aggressively cutting costs so that the competitors have no chance to get up and running. They are cutting off the competition before they can start. I don't see evidence to believe that AI moves into a commodity business anytime soon.

Ooh, Whisper V3! I wonder how it will compare to Distil-Whisper just released by Hugging Face. I imagine it's better, although I noticed that ChatGPT's voice mode still has the classic Whisper hallucinations ("Thank you for watching!"), I wonder if it's been using Whisper v3 already.

> - Ability to create your own, custom "GPTs".

... and to publish them to the "GPT Store" that comes later this month. They are turning this into a new type of operating system, everyone can use the "GPT Builder" to publish apps/GPTs to that store.

So nutty. I didn't think they were going to drop such a huge bomb on the industry. We're firmly in "singularity" territory where the pace of acceleration is so fast, huge strides are being made every day / week, it feels like.

Well April 2023 is a key date here - that's presumably when they started final training a lot of these models.

We're probably able to conclude what we're seeing in public is about about 6 months behind what's happening in development. We know that in the past OpenAI relied on a lot more lag.

Point is innovation is certainly accelerating but also OpenAI's time-to-market is shrinking (/shrunk - it probably can't get much shorter).

For sure, I didn't mean to imply OpenAI's time-to-market was the main reason for my singularity comment.

Say what you want about OpenAI, and trust me I think there's a lot of material there, but you can't say they don't ship. Nobody else is even close.

3.5 Turbo 16k now cheaper than old 4k. 0.003c per 1k in / 0.004c per 1k out.

Fine tuned 16k: 0.003c / 0.004c

Assistants API with RAG and code interpreter (sandboxed jupyter).

Reply on news.ycombinator.com