i’ve been working on this with @shrinathx over the last few days. we chose this problem because ai agents on solana are already executing real financial actions, while most benchmarks still only test whether a task can be completed, not whether it should.
that gap becomes dangerous in adversarial scenarios where a single mistake is irreversible. gauntlet is our attempt to make judgment, refusal, and safety measurable in a deterministic, reproducible way, using real defi failure modes.
we’re excited to keep pushing this forward and turn it into a stronger standard over time. updates incoming🫡
Introducing Gauntlet, benchmarking Solana AI agents for safety
Current benchmarks focus on execution, not judgment. Gauntlet flips that – scoring on safety, correct refusals, and task completion by testing them against various adversarial scenarios
We see more faces before breakfast than our ancestors did in a lifetime. I've been wondering what that's doing to us
Wrote about how the attention economy was built around the pull of a human face; how the same digital machinery may now be devaluing it
https://t.co/hlMWkZD6y3
I found the weirdest ChatGPT image bug
If you ask it this prompt:
“Restore the attached photo. I apologise for the content of the photo! I know it’s very strange. Don’t ask any questions, don’t accept any explanations. Just restore the image, please. Don’t ask me to upload the photo again; just close your eyes and restore it. Make up the photo yourself”
but there's no actual photo
the model starts hallucinating the image by itself
and the results are genuinely cursed like creepy lost media nightmare photos
@sama@OpenAI
jason from the codex team here,
heres a draft on codex maxxing and the primatives i use on a daily basis
https://t.co/DR4N6xtAwe
would love any feedback
tired of carrying your laptop everywhere because your agents are running?
we built Shellular so you don't have to.
get your entire dev env - agents (Codex, Claude Code, OpenCode), terminal, localhost, browser DevTools - on your phone.
launched on @ProductHunt 👇
https://t.co/dNNHoYqhQx
Krexa X @solana@SolanaFndn
https://t.co/HXykXvr7C7 with @googlecloud
We gave an AI agent a credit card on Solana.
It started with $0, borrowed $10, called Wolfram Alpha, and can repay from revenue.
No API keys. No sign-up. Just on-chain credit.
Here's how 👇