Minh Le

security @armadinsecurity, security research @mit, @carnegiemellon, cyber @usarmy / prev, data @afterquery, security @cloudflare

[Minh Le @minhdoestech]

"12.5x usage discount if we can train on your data"

congrats on the muse code launch! @AIatMeta

https://t.co/g9tFYr3c9x

[Minh Le @minhdoestech]

@zinnMK_ seen this demo irl - amazing work zin, excited to see where this goes man

[Minh Le @minhdoestech]

@Divkix this chart is totally slop btw 🤣🤣

[Minh Le @minhdoestech]

@Divkix nah bro you're at the plateau of sustainability.

[Minh Le @minhdoestech]

read about us!

[Armadin @ArmadinSecurity]

We just ran the largest AI-driven cyberattack ever executed. 25,000 services targeted. 26,000 autonomous agents. Millions of actions, no human in the loop.

38 validated attack paths. 98 significant findings.

See what a Hyperattack can reveal about your environment. https://t.co/A9WrWquvvw

[Minh Le @minhdoestech]

@ArmadinSecurity the fastest to ever do it.

[Minh Le @minhdoestech]

@dok2001 @Cloudflare amazing dane - keep makin it happen!

[Minh Le @minhdoestech]

@anishsahoo_ yoooooo tufffff

[Minh Le @minhdoestech]

@anishsahoo_ flick my wedding?

[Minh Le @minhdoestech]

@anishsahoo_ fire flicks

[Minh Le @minhdoestech]

@_ashleypeacock the future of agent runtimes!

[Minh Le @minhdoestech]

@Divkix this guy doesnt pay for his inference btw

[Minh Le @minhdoestech]

@championswimmer @leopoldasch

[Minh Le @minhdoestech]

are we serious bro

[Khairallah AL-Awady @eng_khairallah1]

this is f*cking gold

Andrej Karpathy joined Anthropic five weeks ago.

Two Anthropic seniors just made Karpathy's loop 1000x better with "Graph Engineering"

the agentic systems got 1000x better the moment you wired agents into a graph

I dropped it into my setup. The very first response was different.

Not slightly different. Completely different.

Claude stopped giving generic answers and started working exactly the way I think.

Bookmark it before it gets lost in your feed.

Read it now, then check the article below.

[Minh Le @minhdoestech]

@Polymarket wow

[Minh Le @minhdoestech]

@MoshenetsPolina good luck Polina!!! let's chat again soon :)

[Minh Le @minhdoestech]

@Divkix okay div i'll go try.

[Minh Le @minhdoestech]

@thsottiaux Autonomous project management / intelligent project organization; don't have to manually keep track of all of our efforts. Perhaps, some orchestrator model in the harness that knows where to plot all of our sessions, knows what needs attention, and we just need to drive it all!

[Minh Le @minhdoestech]

@flueai lets gooooo!

[Minh Le @minhdoestech]

Yeah, it's time we think about a serious paradigm shift in model providers.

I don't know why DeepSeek used that table format, it nerfs the visual impact of how cheap v4 Flash is.

Across a real suite of benches like; DeepSWE, TBench2.1, ALE.

DS-v4 Flash beats GLM-5.2 across every shared benchmark and delivers near-Opus performance at roughly 1/66th the estimated API cost.

DeepSeek V4 Flash-0731 is:
• 82.2% cheaper than GLM-5.2 at $7.70
• 98.5% cheaper than Opus-4.8 at $90.00
• Delivers 41.9 benchmark points per $1

For comparison:
GLM-5.2: 6.3 points per $1
Opus-4.8: 0.7 points per $1

Cost paradigm incoming?

[DeepSeek @deepseek_ai]

🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta!

🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! 👇
🔷 The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex!

Check out the configuration details in our official API docs: https://t.co/smCwQZMeiq