simon,
@simon@simonwillison.net avatar

At OpenAI DevDay, a ton of new stuff:

  • gpt-4-turbo (128,000 tokens, faster, cheaper)
  • Ability to pass a seed and get repeatable outputs (maybe we can finally write some automated tests!)
  • Can actually produce valid JSON now - plus call multiple functions in one response
  • Text to speech API (6 voices)
  • Whisper v3 (open model release)
  • Fine tuning for gpt-3.5-turbo 16k (and preview access for GPT-4)
bocytko,
@bocytko@hachyderm.io avatar

@simon Assistant API is an easy way to productize what was already possible through tuning of system message and choosing a set of plugins?

simon,
@simon@simonwillison.net avatar

@bocytko yeah, it's a effectively a quick way to bundle a system prompt and some plugins (and RAG documents) together in one named entity

bocytko,
@bocytko@hachyderm.io avatar

@simon nice. It makes a bunch of the orchestration code needed before completely obsolete.

miki,
@miki@dragonscave.space avatar

@simon Nothing on programmatic access to the vision models so far?

matt,

@simon Do we know yet how the TTS compares to Eleven Labs? Several of us in the blind community started playing around with Eleven Labs, for non-screen-reader use cases, several months ago.

simon,
@simon@simonwillison.net avatar

... and "Copyright Shield" - similar to existing programs from Microsoft and Adobe, OpenAI will now step up with legal assistant if you get sued over copyright issues from output produced using their models

simon,
@simon@simonwillison.net avatar

Details of some of the releases on their blog: https://openai.com/blog/new-models-and-developer-products-announced-at-devday

I'm not yet able to access the model - I get this error:

Error: The model gpt-4-1106-preview does not exist or you do not have access to it. Learn more: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4.

JesseSkinner,
@JesseSkinner@toot.cafe avatar

@simon "We’ll begin rolling out new features to OpenAI customers starting at 1pm PT today." so try again in 2.5 hours I guess?

nico,
@nico@lepoulsdumonde.com avatar

@simon yep, listing the available models doesn't show either the new gpt4 or dall-e-3 yet

nico,
@nico@lepoulsdumonde.com avatar

@simon "We’ll begin rolling out new features to OpenAI customers starting at 1pm PT today." In a couple hours then !

neil,
@neil@mastodon.nz avatar

@simon That new model just worked for me on an API call (I'm a GPT Plus subscriber). 😊

"created": 1699306678,
"id": "chatcmpl-xxx",
"model": "gpt-4-1106-preview",
"object": "chat.completion",
"system_fingerprint": "fxxx",
"usage": {
"completion_tokens": 291,
"prompt_tokens": 3373,
"total_tokens": 3664

kcivey,
@kcivey@mastodon.social avatar

@simon "Legal Assistant" being their LLM-based lawyer emulation?

  • All
  • Subscribed
  • Moderated
  • Favorites
  • random
  • DreamBathrooms
  • magazineikmin
  • cubers
  • InstantRegret
  • cisconetworking
  • Youngstown
  • vwfavf
  • slotface
  • Durango
  • rosin
  • everett
  • kavyap
  • thenastyranch
  • mdbf
  • megavids
  • khanakhh
  • modclub
  • tester
  • ethstaker
  • osvaldo12
  • GTA5RPClips
  • ngwrru68w68
  • Leos
  • anitta
  • tacticalgear
  • normalnudes
  • provamag3
  • JUstTest
  • All magazines