Join us for OpenAI’s first developer conference on November 6 in San Francisco | Hacker News Reader

Join us for OpenAI’s first developer conference on November 6 in San Francisco

openai.com

67 pointsdavidbarker26 comments

ah amazing! I was somewhat running our AI Eng Summit because I perceived that there was a gap in engineer/developer-oriented conferences in AI (a lot of AI conferences are still TED style talking head stuff rather than builder oriented). definitely keen to attend in person, they probably have something big to launch if they're doing an event.

i'm guessing at minimum GPT4 Vision API? with a possible official Advanced Data Analysis API kicker?

I wonder how they're counting those two million developers?

Realistically, probably number of accounts that requested API keys. A lot of people (YouTube, blogs, etc) suggested that as an alternative to paying a higher price for ChatGPT.

Probably counting the number they kicked from their discord server to require ChatGPT driven verification.

They asked ChatGPT and that's what it hallucinated :)

I was writing that maybe they're using one of their AI models to count, and we all know how they count.

Certainly meaning that a developer version of GPT-3 is going to be open-sourced since it has been beaten and over-taken by other open source models.

I would be terrified of building on top of openai’s api. At any moment they can just cut off access or mark a model legacy (or change its behaviour but keep the same name) and all my work is bust.

You should build with the mindset that you will eventually swap out ChatGPT if needed. LLM options will only improve (Llama 2 comes to mind, but lots of folks in the race). As long as you’re prepared to decouple, you won’t get caught flat footed.

I mean, you are somewhat continuously (in an automated/programatic manner) running evaluations of other options against a benchmarking system for your use case, no? If not, consider if/when resourcing permits. This not only prepares you for decoupling, but also allows you to level up if OpenAI's offerings are no longer the most optimal for LLM consumption.

Go build.

I don't see what's particularly terrifying about building on OpenAI.

Yes, services can deprecate or shut down. That's true for any old service and also true for new stuff. They might swap things around but obviously OpenAI will not die of attrition any time soon.

Secondly, it's not like OpenAI has a complicated API. Far from it. Any other LLM API is basically the same, just because the components are so straight forward. So, given enough capability from competitors, as far as SaaS lock-ins go, this is fairly cushion-y.

What remains is a testament of how singularly potent you think the product might be and how much it would advance your stuff beyond what any other product can do. That strikes me as a rather questionable reason to not make use of it, if getting stuff done, efficiently, is what you want.

You could look at it as the chat requests being the API, which is simple yes, but in a way the prompts you would write could also be considered the API, in which case it is the most complex API ever. People aren't worried about having to change the first one, it's the prompting needing to change/ model not having the right capabilities.

Is that any different than building on top of some other api?

I guess it boils down to how much you trust the leadership that owns the API. In this case it seems like the leadership suffers from paranoid delusions, and may therefore not be trustworthy in the long term. You will have to make that judgment for yourself though. https://www.wired.com/story/what-openai-really-wants/

We've been live with them since May. Their service has only gotten better for us over time - faster and no drop in accuracy. For us, the proof is in the pudding - they take being a platform seriously.

We're not wed to them, and actually need to go a different route for some of our customers due to needing to sign a DPA, but the decision to go with openai has been extremely worth it.

You should be more afraid of how often it errors and times out.

Ehhh, this was a problem about 3 months ago. We used to see it regularly. Now we haven't gotten a single "429 model overloaded" error in the past month. Request volume is a little higher for us too.

May I ask what your app does or what you're using gpt for?

Natural language data querying: https://docs.honeycomb.io/working-with-your-data/query-assis...

User input --> bunch of contextual stuff --> API call to openai (could be any model though) --> query executes against user data and they get a result

> “We’re looking forward to showing our latest work to enable developers to build new things,” said Sam Altman, CEO of OpenAI.

Let the "GPT-5 is coming" news cycle begin...

I think it might be announced at the event but not ready just yet.

They've got so much stuff to figure out not only for the next release but also future releases. They've already dialed back so many things this year, including fighting back against jailbreaks, removing the ability to look up people (unless they're prominent in "media"), and adding obnoxious amounts of disclaimers† to every other query (which can now be circumvented with the custom instructions).

†: for ChatGPT UI, the API is fine

Seems like a low effort after thought. Knee jerk reaction from Google Cloud Next?

What makes you say that? There doesn't seem to be a ton of info on that page about _what_ it is. Certainly not enough to call it an after though.

Everything in your comment and the fact that it is two months away. Seems like something they cooked up within the last couple of days. No schedule, no details at all. This isn't how you get folks to attend something, especially something that soon. They should at least put free drinks...

These events will be categorized as moments in history.

I doubt that, it's just marketing.

Aren't apple events where they launch new products just marketing ?

Also Caesar pausing before crossing the Rubicon and grandma talking up the family sauce’s history.

Reply on news.ycombinator.com