Greg Pstrucha
@grichadev
ai @ sentry. undistinguished engineer.
prev robinhood/facebook
521 Following 1.3K Followers
yes you should delete all unit tests
im saying that completely seriously and genuinely, but i thought they’ve discontinued working on antigravity
Antigravity 2.0 now features a dedicated planning mode, just like the Antigravity CLI.
Type /plan and the agent steps back to think through the task, and conduct extensive research, before generating an implementation plan for your review. The agent will ask for your approval before it begins executing on the task.
You can also ask for a plan naturally in your prompt, for a lighter version of a plan.
Show more
slop causes laziness causes more slop
now that i think about it “jev” as name is very sneakily fitting
@typesafeai did u plan this
me explaining why we can't run a classifier on every event
his shitposting game is stronger than i'd expect
I look at it differently
I think most people I knew got into the industry because they really liked coding. Now that AI is in the process of affecting every part of that industry its not the ego that is being shattered but process of the craft we got paid to do and loved is being shattered.
But of course I think there is many things to be excited about and I do think that even though the old way is going to fade out over the years for the majority of programmers, there is still so much amazing things you can do.
Not a doomer
Show more
i've been trying to run warden security benchmarks against 4.7 since yesterday and it hasn't been great. the last grok model that was absolutely crushing it in vulnerability scanning is 4.5 and we still use it as base driver.
both 4.6 and 4.7 disappointed, were spinning for two long, missing or "disproving" some real results and costing way too much.
original numbers are here (although we didn't add 4.7 yet).
Show more
"while retaining code quality"
Just one engineer and a team of agents ported the GitHub Copilot agent runtime to Rust, shipping 800,000 lines of production code while retaining code quality. 🚀
How it worked 👇
Show more
i'm gonna slowly inject @sentry theme to all coding harnesses.
hey @thsottiaux look how pretty. idk, maybe should be in the picker
ok @Muse is good, switched over fully to it
this is very cool
what if we could simulate jev with an oss language model and token probabilities?
we break down how jev works and shipped a demo running with DiffusionGemma on @CloudflareDev workers ai:
Show more
what dreamforce taught dario about b2b saas sales
We’re partnering with Accenture on independent evaluation of frontier AI—part of our recent commitment to embed evaluators at Anthropic. Both we and Accenture expect to invest at least $1 billion to build capacity in this area over the next five years.
Show more
chief proxy officer
I wasn't able to get it to a 100%, but I found that if I run it as a confidence check, jev is pretty consistently flagging *all* my eval'ed issue and rejecting red herrings at 0.65 confidence or over. i can still send the remaining issues for double check to a larger model and reap the benefits of it being cheap.
more interestingly, it being confidence level is a lever i can pull, if I want my pipeline to catch things more conservatively at the trade-off of slightly larger cost (for a second pass), that's very configurable and valuable to me.
Show more
I wasn't able to get it to a 100%, but I found that if I run it as a confidence check, jev is pretty consistently flagging *all* my eval'ed issue and rejecting red herrings at 0.65 confidence or over. i can still send the remaining issues for double check to a larger model and reap the benefits of it being cheap.
more interestingly, it being confidence level is a lever i can pull, if I want my pipeline to catch things more conservatively at the trade-off of slightly larger cost (for a second pass), that's very configurable and valuable to me.
Show more
did you know that @Cloudflare is 🐐ed?
update: jev is live on @cloudflare ai gateway
🪅
please get them on i have much more experiments to run @typesafeai
do i know anyone at @typesafeai? would love to get jev on @cloudflare ai gateway
these are results of using jev on one of our security pipelines. we already have to use smaller (and dumber) models to make it economic and this model does it over 5x cheaper, faster and while maintaining much higher accuracy.
i don't normally hype over new model releases but that's the first one that actually impressed me. good job @typesafeai
Show more
