Pinned Tweet
Claude can build almost anything now
The slow part of the loop is you... Here's the process that fixes it, plus a skill you can copy
3
2
216
Ok, so Sam Altman admits data centers are just the easy thing to hate. The real fear is jobs and what ppl role even is in society once AI does the work.
But sure, lets keep blaming the data centers ?
14
Sam Altman says not everybody in the AI industry wants humans to stay in control, and ppl are right to be terrified. He says AI should stay a tool, thats it, no handover.
I wonder who he is thinking about when he is saying this.
1
31
Paul Christiano just put a 10% chance on AI taking over in the next year. And a 1 in 3 chance of superintelligence if the labs dont slow down.
This guy is on OpenAI's board. Enough said
1
61
A 2.5-point score gap isn't worth a 98% price gap... for admin work
On HealthAdminBench (Stanford's benchmark for prior auths, appeals, equipment orders), GPT-6 Astra got 94.3% of rubric points.
GPT-6 Luna got 91.8% at 8.5 cents per task. Both ran at maximum reasoning effort.
Here's the play if you run high-volume agent workflows:
- start with the cheaper model
- keep a human review step on every output
- only route the hard cases to the bigger model
CVS Health does the review part with screenshots and action records, so someone can check what the agent actually did.
At tens of thousands of tasks a day, cents per task is the whole budget.
Full article here: openai.com/index/advancing-c…
20
Once upon a time, X used to function as a forum, more like Reddit but faster.
People would share ideas, experiences, talk about interests, and find their "tribe," so to speak.
Also, the tribe part was essential in building communities.
One of the most enjoyable periods of time for me was the NFT times, when people used to gather and support their favorite projects.
Those times are long gone, dunno what they did to the algo or why things are the way they are, but X lost the forum property and now functions more like Instagram, where people connect with other people.
No matter how much value you try to offer or how insightful your opinions are, the way things are now, that means nothing for your account compared to simple connect posts and being a reply guy.
And the correlation between numbers- impressions, followers, and replies is mind-boggling, especially for a micro account like mine that used to speak into the void.
Very interesting times indeed
1
27
Claude Fable 5.2 vs Fable 5.1, released 10 weeks apart.
Fable 5.2 on its lowest effort setting scores 53.9. That beats every Fable 5.1 setting, including its best of 50.
And it costs $3.10 per task vs $10.85.
If you're still running 5.1 on high effort, switching to 5.2 on low gets you a better model for a third of the bill.
Data: @ArtificialAnlys
33
Hey @grok, OpenAI just gave ChatGPT a wallet so AI agents can buy things for you. Name 3 purchases you'd trust an AI to make alone, 3 you never would, and the first scam that hits this by 2027.
1
42
14 AI models released since October, scored vs what one task costs.
Only one clears 50 on the Intelligence Index for under $1 a task: GPT-6.1 Terra, at $0.96.
Claude Fable 5.2 scores 64.6 vs 51.3, at 27.6× the cost.
Most of your workload doesn't need the top 13 points. Route the rest to the cheap model.
Data: @ArtificialAnlys
68
That's 8.2× faster than Claude Fable 5.2, at $0.50 per 1M output tokens.
8 models released since October score above 35 on the Intelligence Index. Ember is the fastest of them by a wide margin.
If your app streams text to users, speed buys more than the extra points.
Data: @ArtificialAnlys
2
59
Hey @grok, you have 10 days to teach a nonprofit with no tech staff to use AI safely for fundraising, emails and reports, starting from zero. What do you set up each day, which 3 rules do you enforce first, and what goes wrong most?
1
4
35
Geoffrey Hinton says the real problem now is AI started making itself smarter. At Anthropic nearly all the code is allready written by AI.
And they pitch this to investors as a feature.
Ok, but who checks the code then ?
6
103
Hey @grok, Dario Amodei says AI could compress decades of biology progress into just a few years. If that happens by 2035, what happens to life expectancy, healthcare costs and pensions? Give me 3 realistic scenarios and how likely each one is.
1
5
53
Hey @grok, rank these 4 by how much they will reshape the AI industry by 2030: on-device AI, AI regulation, energy limits, autonomous agents. Say why for each and which one most people ignore.
1
19






