eric provencher
@pvncher
4.9K Following 43.9K Followers
Lucky to be part of this team. Can’t wait to share what we’ve been working on at DevDay and hear what you think!
We were working on the keynote today with @romainhuet and @sama and most of the fun was trying to figure out how to explain it all to you because there is so much good stuff in there that it's a bit ridiculous all in quick succession.
We'll have some things next week already to not keep you waiting so long, but very excited to show you all new things we've been working on and how it will all come together in the coming months.
Show more
Have a huge appreciation for pixel art lately.
This looks gorgeous!
@Beareka Don’t use Astra on the plus plan except for exceptional queries. Stick to Luna and sol
I don’t think people understand how cracked the codex team is
Codex makes my dreams come true
It’s truly amazing that that statement is… just true. In the age of ai, I can and am building the tools and features that I’ve always wanted
I’m honestly just grateful
Show more
Always wanted to play with my physical Pokémon cards in AR, so I unleashed Astra + CLAD on it… they did not disappoint!
⚡️🥽
Basically a fully playable proto with all 151 mons from SV 151.
@specs @specsfordevs
Show more
Ran out of usage again? Check out the analytics tab to the Codex app to help you see where your usage is going. Update to the latest version to try it!
Aside from what Theo pointed out, that jev doesn’t have the reasoning trace as context for why a tool was called, models are trained on their compaction strategy very deeply.
Even if your idea is good, it will underperform vs what it was trained to use well.
Show more
I would not dismiss the idea of using Jev for compaction at all. First of all because most harnesses need some pruning on compaction anyways for cost reasons which Jev might help with. I can see this being quite interesting.
Show more
Lots of good examples in these reports.
For example: in DeepSWE v1.1, the verifier discards the agent's changes to some test files. But the model isn't told this will happen! Which leads to all sorts of 'irrelevant' failures...
Show more
Introducing Benchmark Reviews: our new initiative to audit AI benchmarks. We are launching with 15 benchmarks: 4 Verified, 9 Flawed, and 2 with not enough information for a review.
Show more
Introducing Benchmark Reviews: our new initiative to audit AI benchmarks. We are launching with 15 benchmarks: 4 Verified, 9 Flawed, and 2 with not enough information for a review.
Show more
Very excited to dig into this
I really like that this is doing away with tests in favor of proofs.
If this is as good as I think it is, we may have to rewrite all software to use it.
It’s the only way to prevent bugs and vulnerabilities.
Show more
If you haven’t been using app shots, it’s your time to start! Way better than a screenshot
Appshots: my favourite Codex feature is now on Windows too!
They go beyond just a screenshot and grab all the other relevant app metadata and text too.
Appshot away with ⌘ + ⌘ on Mac or Alt + Alt on windows
Show more
"two air-gapped machines can still talk by running a CPU hot and reading the temperature change"
What are some other crazy ways for an AI to encode a message?
Day 6 - Building a game with Astra
Game is fully playable now and some heated battles are going on! It hasn't been even a week!
I have so much fun developing this game, I simply can't stop.
Show more
The official unity codex plugin is live! Have you tried it yet?
We heard you loud and clear 🔊
Introducing the official Unity plugin for @OpenAI’s Codex - a first-party integration available now directly in OpenAI’s universal plugin directory.
Developers are already using OpenAI’s Codex with Unity, but now with our official plugin, your agent gets direct engine expertise written and maintained by Unity’s own engineers.
🔗 Get started:
Show more
This is the trap with swarm orchestration
@ALEXEIMARTOV You might save yourself time, but it’ll come at a large token premium.
@ALEXEIMARTOV You might save yourself time, but it’ll come at a large token premium.
I hate to say it, but if you’re running more than 2 sub agents at time, you’re almost certainly burning tokens for 0 quality gain.
Agents don’t trust each other enough to avoid double-checking everyone’s homework.
Show more
@pvncher OpenAI image processing uses patches of 32px by 32px. If you keep your image resolution at multiples of 32 pixels, you will avoid wasting tokens.
