Kevin Roose
@kevinroose
Followers
167K
Following
14K
Media
155
Statuses
887
NYT tech columnist, Hard Fork co-host, high-perplexity language model. Writing a book about AGI.
SF
Joined January 2009
Some news: I'm writing a book about AGI! I've spent the past few years hanging out in San Francisco's AGI scene — the small cluster of people working to build human-level AI, or trying to stop it from being built, or living as if it's already here. I've heard a lot of great
72
35
504
“Meta wasting tens of billions of dollars on the metaverse was bad” my brother in Christ, can you imagine if Meta had not gotten distracted by a shiny object during the LLM takeoff and was still a top-2 AI lab? the metaverse might have saved humanity
17
20
662
It's Hard Fork Friday! This week on the show: • OpenAI declares code red! • Our hands-on impressions of Gemini 3 and Opus 4.5 • And the Hard Fork Review of Slop returns, with a fake Christmas market, a fake Whirlpool ad, and Bird Game 3, the fake video game sweeping TikTok
2
0
18
One thing in life is certain: you’re going to die. Oddly enough, that fact can be liberating. Remembering that life has an endpoint helps you live it so that it has a point—a direction, a storyline, something more than a string of disconnected
9
15
328
i have stopped being shocked when AI pundits, people who think and talk about AI for a living, people who are *writing and sponsoring AI legislation* admit that they never use it, because it happens so often. but it is shocking!
4
3
57
had to check the date on this one
Sen. Josh Hawley, one of the biggest AI critics in the Senate, told me this AM that he recently decided to try out ChatGPT. He said he asked a "very nerdy historical question" about the "Puritans in the 1630s." "I will say, it returned a lot of good information."
4
7
119
It's the most fun I've had with a model since Sonnet 3.5 (new), the OG god model. Gemini 3 is also remarkable, for different kinds of tasks. My working heuristic is "Gemini 3 when I want answers, Opus 4.5 when I want taste."
2
3
150
Now you run Hollywood. Show them how it's done. Wishlist 'Hollywood Animal' on Steam NOW! https://t.co/aGUcTYFd1y
14
36
476
Claude Opus 4.5 is a remarkable model for writing, brainstorming, and giving feedback on written work. It's also fun to talk to, and seems almost anti-engagementmaxxed. (The other night I was hitting it with stupid questions at 1 am and it said "Kevin, go to bed.")
47
31
988
Podcast hosts: never forget to ask your silly joke questions, like "should there be a Polymarket-for-kids inside Roblox?" You never know when you'll get a sincere, enthusiastic yes.
25
122
1K
Roblox is the largest kids' gaming platform in the world, and it's facing 20+ lawsuits, and multiple state investigations over issues related to child safety. Parents should know how its CEO responds when asked about them!
12
53
790
To be clear, Roblox pitched *us* on interviewing their CEO about child safety. We don't ambush people. But he seemed totally caught off-guard by the questions, and got angrier the more we asked.
35
175
2K
This week's Hard Fork might be the craziest interview we've ever done. We asked the CEO of Roblox about the child safety issues that have been plaguing the platform. His response was...well, just listen. https://t.co/VBjvLmPYP7
nytimes.com
This week the company announced that it would require users to undergo an A.I.-powered age estimation process in order to chat with others on the platform. Will this change reassure parents?
62
191
1K
Gemini 3 emergency pod! We talked to @demishassabis and @joshwoodward of Google DeepMind about America's next top model
nytimes.com
Maybe more than other model releases, this one seems to have the attention of Google’s competitors. Will it put the company at the top of the A.I. leaderboard?
26
45
433
It's Hard Fork Friday! This week on the show: • We're putting data centers in SPACE • Former WH policy advisor @deanwball tells us what MAGA thinks of AGI • And @HistoryGPT explains how a new Google model (Gemini 3?) one-shotted his history mystery https://t.co/SRIC3RdA8Q
nytimes.com
“As you may have noticed, it is not easy to build data centers here on Earth.”
2
7
26
This is correct, and also true of every recent AI paper (the METR slow-down study, the Apple reasoning one) that casts doubt on AI's effectiveness. People are desperate to prove that LLMs don't work, aren't useful, etc. and don't really care how good the studies are.
I think folks who know better, esp on twitter, are still underrating the extreme impact the MIT paper had about 95% of AI projects failing within enterprises. I keep hearing it over and over and over again.
55
31
381
Apparently the correct way to write 5,000 words a day is by drinking two Celsiuses and listening to Somebody Told Me 150 times, who knew
4
2
48
After I heard Michael Lewis say on a podcast that he listened to "Let It Go" from Frozen on repeat while he wrote books, I tried it. (Not with Let It Go, but the same song over and over again until it becomes white noise.) Instant 20% productivity boost. Thank you Michael Lewis.
22
6
159
the most powerful people in the world are the business schools and consulting firms running studies of whether AI speeds up white-collar workers or slows them down. trillions of dollars, the fate of nations hangs on some small-n RCTs and self-report surveys about how often
13
25
284
PewDiePie becoming a cracked AI guy was not on my 2025 bingo card but I don't hate it
PewDiePie just vibe-coded his own Chat UI, built an army of chatbots for majority voting and gave them all RAG, DeepResearch and audio output naturally, he only uses chinese Qwen models and runs them on his local PC with 8x modded chinese 48GB 4090s and 2x RTX 4000 Ada his army
2
4
92
It's Hard Fork Friday! This week on the show: - Character AI is backing off chatbots for minors - Casey shows me my Grokipedia page - And the great @ajjacobs describes the harrowing 48 hours he spent with no AI at all (it involved foraging) https://t.co/bKowL3P6m0
nytimes.com
“We are living through a dramatic contraction in the access that teenagers have to technology online.”
3
0
13