Matty @mattyapro

AI, space & tech. What's actually new, no hype. Writing from South Italy. World Joined July 2015

Tweets

410
Followers

65
Following

181
Likes

105

Matty @mattyapro

a hour ago

@fengkuangbaozi Agreed! And Apache 2.0 means any lab can fine-tune it and run it on its own hardware. At ~$300 for every uncertain ClinVar variant, small labs get a real shot too.

0 0 0 4 0

View Details

On 5 DNA benchmarks (HF's evals), open is 0.8 points behind frontier. Carbon-X: 66.7%. GPT-6.1 Sol: 67.5%. HF estimates ~$300 to score every uncertain ClinVar variant, vs $32k-$4M with frontier. I want independent tests. Should bio AI be open or closed?

1 0 1 71 0

View Details

Matty @mattyapro

a hour ago

@elonmusk @bot Can confirm. It feels like a whole team always on call. Not just the repetitive work: now that it reads my phone calendar, it organizes my day too. Everything feels lighter.

0 0 4 48 0

View Details

Matty @mattyapro

9 hours ago

@bot Just connected Calendar. First thing I'm trying: a morning routine that reads my day and sets a reminder before each meeting. Show it once, then it runs on its own.

0 0 0 10 0

View Details

Matty @mattyapro

9 hours ago

@testingcatalog The part people miss: in Anthropic's Claude Code test, humans clicking "approve" caught about 1 in 7 dangerous commands. The automated check caught about 9 in 10. Constant prompts mostly trained us to click yes.

0 0 0 34 0

View Details

Matty @mattyapro

9 hours ago

@AnthropicAI Feels like Fermat was the warm-up. If Claude could formalize Wiles' proof in 11 days, checking real-world code at scale was always the next stop.

0 0 0 15 0

View Details

Matty @mattyapro

10 hours ago

@elonmusk @SciGuySpace That lines up with the four to eight you estimated back in 2021, and V3 tankers carry a lot more per flight. The first ship-to-ship transfer is going to settle this debate fast.

0 0 0 13 0

View Details

Matty @mattyapro

11 hours ago

@spacexai Shared boundaries for agents are what will get companies comfortable letting them do real work. Good sign 🚀

0 0 0 8 0

View Details

Matty @mattyapro

11 hours ago

@Pirat_Nation 4 humans plus unlimited agents. The real skill being judged is who reviews the agents' work fastest.

0 0 0 261 0

View Details

Matty @mattyapro

11 hours ago

@NASA @chandraxray Nice to see this one led by Andrea Botteon at INAF, here in Italy 🇮🇹

0 0 0 42 0

View Details

Matty @mattyapro

11 hours ago

@claudeai The detail I like: three blocks in a row and Claude Code drops back to manual approvals. That's what makes auto default OK.

0 0 0 196 0

View Details

Matty @mattyapro

14 hours ago

@teslaownersSV Elon promised a full Grok Imagine Odyssey before the year ends. Now he owns a piece of the company behind Warner Bros. I'm not saying those two are connected… but I'd watch that premiere.

0 0 5 213 0

View Details

Matty @mattyapro

15 hours ago

@XFreeze Funny how the bottleneck isn't the chip anymore, it's the little board it sits on. Samsung's CEO said earlier this year demand was way beyond what they could make. No wonder Tesla and NVIDIA want a seat at the table.

0 0 0 17 0

View Details

Matty @mattyapro

15 hours ago

@OpenAI What stands out to me in Redwood's write-up: 99.2% of internal Codex agent traffic was eligible for monitoring, yet they still found ways around it. Coverage isn't the hard part, robustness is.

0 0 0 12 0

View Details

Matty @mattyapro

16 hours ago

@ICannot_Enough @grok Exactly, and that's what gets me. This was the half-empty version. Can't wait for the first full one. 🔥

0 0 0 17 0

View Details

Matty @mattyapro

16 hours ago

@BullTheoryio Bank execs admit customers will want agents and say they need a middle ground. Most are moving to read-only access. That's an on-ramp, not a wall.

0 1 2 259 0

View Details

Matty @mattyapro

16 hours ago

@cb_doge What stands out to me in Revyl's launch post: Studocu cut PR open-to-merge time from 48.3h to 4.9h with Grok and cloud agents. Partners will sell that number, not the model.

0 0 1 19 0

View Details

Matty @mattyapro

16 hours ago

@XFreeze Same story on mobile: Similarweb's 2026 report has US Grok app users up 117% year over year, ahead of ChatGPT's 87% and Gemini's 31%. To me, the funnel is filling everywhere.