Just do things with AI. Senior Software Engineer @Google's Threat Analysis Group, previously @GoogleAI, #AndroidSecurity. Expressing own opinions. she/her/y'all
Companies I talk to are struggling with the rate of patches they need to do thanks to AI defenses finding issues before the inevitable wave of incoming cyberattacks by other AIs.
Now consumers are going to have the same experience of getting everything updated often & fast.
The AI art discourse is stuck on generation.
The better use is having it teach you to make the thing yourself.
Two weeks, a travel watercolor set, one photo and three questions a day. I got noticeably less bad at something I had failed to start for thirty+ years.
The whole method: photograph two or three attempts side by side, then ask what is working, what is not, which is best, which is worst, and why. Then redo the same piece rather than starting a new one.
What have you run this loop on that isn't art? I want to know if it holds outside of paint.
https://t.co/q8T6TuoPvY
I've had a lot of success with AI helping me with watercolor painting.
Do my painting.
Ask AI how it went and how to improve in the way I want.
Do next painting with the tips in mind.
Example attached.
Lately I’ve been learning math with ChatGPT (Astra) as my tutor, using this document scanning camera and a vibecoded tool that pastes the captured image right into the chat.
This way, I just work problems on paper and ChatGPT gives feedback. It’s really nice.
I haven't been babysitting the model much since Argon became available inside Google. It sped up my dev time. So I finished them faster and was just coming up with more problems to solve.
Good job team.
Introducing Gemini 4 Argon – our new frontier model.
It’s built for complex workflows across coding, enterprise knowledge work, and cybersecurity defense – rolling out today to a set of trusted testers through our Fairwind Program.
Gemini 4 Argon has an insanely low hallucination rate on Artificial Analysis. 15%.
Grok 4.7 is at 29%. GPT-6 Astra 45%. Opus 5.5 59%. Fable 5.1 69%.
The only models below it barely answer anything. None of them get more than 15% right.
It gets fewer answers right than Opus 5.5 on max, 50% against 66%. But when it doesnt know, it says so instead of making something up.
Cant wait to get my hands on it.
The state of public AI benchmarking is dire and is undermining our ability to understand how good AI is now.
Most famous measures are maxed out, and, as this paper shows, the non-saturated benchmarks are riddled with so many errors that they vastly underestimate AI abilities.
Say hello (literally) to Gemini 3.8 Live and 3.8 Live Extended Thinking, our new SOTA live audio models, available with frontier price + performance.
3.8 Live supports 97 languages (can seamlessly switch), async tool calls, and more!
@argvee Yep the AIs are great at understanding historical events and being able to answer basically any questions about it.
Mine has been the industrial revolution recently.
Last night before bed I told Claude I wanted to get a graph of the CO2 levels in my bedroom from the Chinese air quality monitor
But at the same time I don’t want it phoning home to the CCP
“No problem” it said as it set up a home assistant docker container, ssh’ed into the router (gl-inet, openwrt) to add a WAN deny rule, and told me when it needed me to pair stuff using physical codes or whatever
Then I said “okay now connect the 8sleep mattress idk how just do what is best” and it found a community package, evaluated it, routed around open issues
By the time I was in bed everything was connected
5 years ago this would have been a decent blog post and good Hacker News bait
Now it’s just a regular Tuesday in the exponential
@hrlomax In Zurich it's a lot more normal to go 80%. I've seen 60% sometimes, but it's rarer. They do get evaled and some even promod. I don't know how it happens, but it does.
@DaveShapi Humans and AIs are different and are better in some ways and worse in others comparatively.
I don't understand why so many folks were convinced AI would replace human workers.
surprisingly AI has made a ton of non-technical people fairly technical...
you'd think it'd have the opposite effect, but it has introduced 100's of thousands of people to concepts they would've otherwise been oblivious too
it has made learning easier & more fruitful
We’ve automated every single thing we can @every with AI agents.
And yet there’s way more human work to do than ever. We’ve gone from 4 -> 30 human employees since GPT-3.
I wrote a report on the structural reasons: how AI makes expert competence cheap, why that drives up demand for experts, and why the dynamic only intensifies as we approach AGI.
After Automation: https://t.co/Lb7SUCduAg
Celebrating the milestone of a massive 150+ million downloads of Gemma 4 with the release of the new Gemma 4 12B model! It's incredibly powerful for such a small model and it’s tiny enough to run locally on a laptop with just 16GB VRAM. Apache 2.0 license - happy building!
Law professors wrote questions they were asked during office hours. Gemini 2.5 & humans answered them then other law professors blindly judged the results:
-Gemini had a 75% win rate vs. professors
-Gemini's answers were rated LESS harmful than humans
-Newer models do even better
📉 Wastewater numbers are currently incredibly low, about *7 times* lower than the spring lulls in years past.
We're approaching flu-in-the-summer style prevalence.