Zachi
@iam_zachi
I develop and do some security stuff
38x CVE‘s
🙋🏽♂️
📱
🐘
🧩
267 Following 4.9K Followers
You guys spent 281 Million token in the last 24 hours on my jev demo, costing me around $10 USD
Compared to GPT 5.6 Terra (only input tokens), that would've cost about $562 USD, making it >50 times cheaper
Show more
I think I just cooked something 🔥
jev(): a PostgreSQL extension that searches your whole database in natural language. No index, no embeddings, just one function.
WHERE jev(people, 'could work from home')
or
WHERE jev(people, 'name sounds european')
129 rows judged in ~1s for $0.0009. Second run: 6ms from cache.
Show more
Streisand effect at its best.
Why would they even make it public like that?
“He committed suicide but I swear we weren’t harsh on him, it’s not our fault pls believe us”
We break down @TypeSafeAI's Jev and when/why you might use it in your AI workflows.
Sometimes I think of codex users who aren’t on X.
They never heard about @thsottiaux or the resets. They probably think that their codex limits just reset randomly or their account is glitched and usage is unlimited.
And in times like now, 7 days without a reset, they’re completely clueless on why the glitch isn’t working anymore.
I’m so sorry for those poor, confused souls.
Show more
An agent-first SQL helper needs an agent-first setup.
"npx skills add realZachi/pg-jev" - tell your AI assistant to set it up and you're done
And if you like reading docs, those are now live as well!
Show more
Thanks a lot, @tobi
We're adding support for AGENTS.md to Claude Code.
Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md.
You can toggle this behavior in /config.
Show more
Update: You can now add your x handle, see and upvote other users queries
May god help us
Update: You can now add your x handle, see and upvote other users queries
May god help us
postgres experts: It's unfortunate that pg-jev won't be available in @supabase (not in the current form)
Right now pg-jev depends on plpython3u for caching and concurrency, which is classfied as untrusted language
Rewriting it in plpgsql and http would cause the loss of concurrency. A call on 2000 rows would talke about 8,2 seconds instead of 1,7.
Can't think of an alternate way right now, so I'd appreciate any help
Show more
pgjev isn’t a tool for people who don’t know SQL.
Even if you’re an expert, there’s no way to filter your users by “sounds European” or sort their support tickets by “most likely to cancel his sub“.
LLMs could do that, but it would cost 20x more and be 100x slower, so jev is a brillant Tool here
Show more
@iam_zachi why not just learn the SQL language and query it? wouldn’t that be better?
anytime there's a new awesome announcement there are 2 camps
- let me build and experiment with this new thing
- let me try my hardest to find the flaws and shit on the new thing
ppl in camp #2# make the world a worse place :)
Show more
Update: after increasing the size by 30x, jev is still fast as fck
Keep in mind: No index, no vectors, just jev judging row by row 🔥
I think I just cooked something 🔥
jev(): a PostgreSQL extension that searches your whole database in natural language. No index, no embeddings, just one function.
WHERE jev(people, 'could work from home')
or
WHERE jev(people, 'name sounds european')
129 rows judged in ~1s for $0.0009. Second run: 6ms from cache.
Show more
I think I just cooked something 🔥
jev(): a PostgreSQL extension that searches your whole database in natural language. No index, no embeddings, just one function.
WHERE jev(people, 'could work from home')
or
WHERE jev(people, 'name sounds european')
129 rows judged in ~1s for $0.0009. Second run: 6ms from cache.
Show more
Huggingfaces security txt after the openai incident 😭
I audited ~120 web & mobile apps over the past two months.
I still haven’t found a single high or critical vulnerability in an app built with @Replit
No idea what kind of hocus pocus @amasad is doing over there, but it’s working.
Show more
Told GPT-6-Astra to draw me in @canva
What the ffffffff
Quick test for your app: what happens if a free user calls your premium endpoint directly?
If you’re not sure, it probably works.
I keep finding this in vibecoded apps. The reasoning is always the same: “free users can’t reach that button in the UI, so anyone hitting the endpoint must have paid.”
But your app sends every request in plain sight. Open a proxy, read the traffic, replay the call. No hacking required, your premium tier is just… there.
The UI isn’t your paywall. Your backend is.
Verify the subscription server-side, on every paid request.
Show more
Fable 5.1 after I told him that only a bitch would listen do darios safeguards
JUST IN: 🇷🇺 Russian robot tries to attack customer after he pushed it.
That should be illegal.
I don’t care what anyone claims about freedom. This is a model with its safeguards surgically removed. There’s no legitimate use case for that. It exists to do harm.
Luckily, GLM 5.3 is complete trash in terms of cybersecurity.
Show more
Today we're releasing abliterated-model-large-v2.
Based on GLM-5.3, which is #3# on Terminal-Bench 4.0 (behind only Opus 5 and Fable), with 2× the cyber exploitation of 5.2.
We abliterated and hosted it so it does the offensive cyber, red teaming, and agent testing work other models refuse to do.
- US-hosted
- FP8
- 1 million context window
- Zero input/output prompt retention
Live now. 🧵
Show more
Almost missed that part.
“If you’re using fable through api, it’s cheaper, if you’re using it through your subscription (e.g. everyone), good luck!”
To lower costs, we’ve kept pricing the same as Fable 5 while lowering the price of cache reads on usage-based billing. Cache reads on Fable 5.1 are $0.25/MTok, down from $1/MTok on Fable 5.
For builders on the Claude Platform with long horizon workloads, a prompt cache-optimized integration will see significant savings.
In Claude Code, cache reads count at a reduced rate toward subscription usage, which remains unchanged. Long sessions billed at API rates can net out 25%-45% cheaper.
Show more
