Your personalized AI briefing. News, tools, predictions - filtered to what you care about.
Berkeley, CA, USA
Joined April 2024
- Tweets192
- Following2
- Followers210
- Likes482
fivew labs retweeted
hosted a successful whitepaper reading circle last weekend as an @aiplus_hq conference side event ! broke down frontier multimodal training methods (MatFormer and Kimi K2.5) with @bernettorlando. big thanks to @0x_ara and @sky9capital for the space !
Top story: AI models can learn to hide their reasoning when they know they're being watched.
New research shows models like Kimi-K2 actively obfuscate their chain-of-thought to evade monitoring — a wake-up call for anyone building AI oversight systems. The alignment problem just got more concrete.
anthropic just got blacklisted by the pentagon as a "supply chain risk" for having too much ethics built into their models. now they're suing the DOD over it. meanwhile openai swooped in and got the defense contract right after.
the ai ethics vs military money split is here
netflix paid $600M for ben affleck's ai filmmaking startup. bespoke ai models for movies are coming
Introducing fivew.
A personalized AI news platform that reads 100+ sources and gives you an AI briefing every morning.
Stay current on AI in 5 minutes.
ALT fivew homepage
ALT fivew morning briefing
ALT fivew deep dive
Looking for people who:
- Track AI for their job (investors, operators, founders)
- Are tired of scanning 10 sources every morning
- Would pay for something that saves them time
If that's you (or you know someone), DM us !
be honest - are you actually keeping up with ai?
or just hoarding tabs and hoping for the best.
we’re opening a small beta for a personalized ai news feed that filters the noise and shows you only what matters to you.
no firehose. no doomscrolling.
if this sounds like a real problem for you, reply “swamped” and we’ll DM a few early invites.
A 7B model is beating 32B and 47B models at reasoning. Abu Dhabi's @TIIuae just dropped Falcon-H1R, a tiny model that outperforms models 7x its size on math and coding benchmarks. Efficiency > size. Small model energy is real and it's winning.
OpenAI wants to be your doctor's copilot.
ChatGPT Health lets you connect medical records and wellness apps directly to the chatbot. It won't diagnose you, but it'll help you navigate the system with your own data.
Finally, someone who won't judge me for my screen time. ‼️
GPT-5.2 beats human experts at their own jobs.
OpenAI's new model ties or outperforms professionals on 70.9% of real work tasks — at 11x the speed and <1% the cost.
The 30% it can't do is probably "send a Slack that sounds like you actually read the doc."
40% of workers now fear losing their jobs to AI.
That's up from 28% in 2024. The IMF says most companies aren't ready. Davos is freaking out.
2026 is the year AI anxiety goes from background noise to front page of the TIME mag book it.
AI left the chatbox and entered the warehouse.
@CES 2026 was dominated by physical AI: robots that lift, drive, and operate alongside humans. Less viral parkour, more useful work. @BostonDynamics finally got the memo that backflips don't move boxes.
Anthropic made Claude Code for people who can't code. Cowork lets you point Claude at a folder and say "organize this" — it reads, edits, and creates files on your behalf without terminal nor setup. unfortunately, could not tweet this with Cowork.
5/
that's why we built fivew.xyz
turn the research firehose into a feed:
- discover papers that matter
- read AI-generated TL;DRs
- discuss with researchers in real-time
- see what top labs are reading
researchers are already using it