Tracking the AI frontier: models, agents, research, leaks & pre-release signals. I test, investigate and build.
Tips, leaks & collabs → DM
- Tweets 6,096
- Following 1,627
- Followers 10,923
- Likes 116,596
Token Gremlin retweeted
Harry Potter #2 is here.
Another AI safety clown who wants to be the next Jacob Coxon.
🚨WTF. This is concerning.
The AI/SI industry now has another Jacob Coxon.
David Robinson, who recently left OpenAI after 3.5 years, has now laid out WHY he quit, and says OpenAI's culture is BROKEN.
This isn't some random outsider. He oversaw safety reports for 12 frontier AI launches and helped draft OpenAI's current Preparedness Framework.
And then comes the alarming part:
He says he NEVER encountered a colleague with experience running safety-critical systems like aviation, nuclear reactors or systemic finance.
>OpenAI runs in perpetual launch-to-launch sprints
>teams rarely have time to make fundamental organizational changes
>its trial-and-error safety culture "guarantees periodic failures"
>AI systems are already far more capable and dangerous than those built just six months ago
His conclusion: "The time for trial and error is over."
Robinson says frontier AI labs now need to operate more like NUCLEAR POWER PLANTS and aviation, with serious planning, safety-critical expertise and layered redundancies instead of learning only after something goes wrong.
15
3
59
4,629
Token Gremlin retweeted
They’re working on all of these fronts at the same time.
The pause around GPT-6.1 Astra is mostly just what we can see from the outside.
They haven’t actually stopped pushing forward internally, and they’re still clearly ambitious about releasing more models soon while improving the safety systems around them.
1
2
19
1,401
A while ago, I posted that OpenAI was moving at an insane pace internally while also becoming increasingly cautious about the safety risks that came with that acceleration.
Now David Robinson has resigned from OpenAI saying essentially the same thing: that the company’s internal culture has become increasingly difficult to reconcile with the level of risk its current models may represent.
OpenAI has historically operated with a very Silicon Valley mindset:
→ Ship something
→ See what breaks in the real world
→ Fix it
→ Keep moving
So some of the things I was hearing weeks ago, even before DevDay and GPT-6 Sol, suddenly make a lot more sense.
According to people around the company, OpenAI may simply be moving too fast, compressing work and releases that could normally take months into much shorter timelines.
1
2
64
6,202
Final results from my AGI poll, with 2,595 votes:
→ Anthropic: 63% (~1,635 votes)
→ OpenAI: 37% (~960 votes)
Honestly, I expected Anthropic to win this one, but not by this much.
Obviously this is just a poll of my own audience, not some scientific measurement of who is actually closer to AGI.
Still, it’s interesting to see how strongly public perception has shifted toward Anthropic lately.
Who do you guys think is closer to AGI?
I’ve been testing GPT-6 Astra and Opus 5.5 a lot, with some early Fable 5.5 testing too, and honestly I’m leaning more toward Anthropic right now.
But OpenAI’s computer use is still absolutely better.
15
37
3,870
Final results from my OpenAI plan poll, with 3,374 votes:
→ Plus ($20): ~1,552 users
→ Pro ($200): ~877 users
→ Pro ($100): ~810 users
→ ProMax ($500): ~135 users
If these responses are actually accurate, I’m genuinely shocked that around 135 people in my audience are casually paying $500 a month for OpenAI.
Apparently I have way more rich people following me than I realized. LMAO
24
3
215
8,999
Do you think Claude is conscious, alive, and possibly a god?
19%
Yes
40%
No
9%
Never gonna happen
32%
Dario is delusional
990 votes • 17 hours
48
1
30
4,406
Token Gremlin retweeted
Holy shit they wanted Claude to become the new Jesus and the Church refused
NEW: According to a bombshell report in the New York Times, Anthropic co-founder Chris Olah threatened to walk out of Pope Leo XIV’s AI encyclical launch in May because the pope rejected the idea that machines can be conscious.
Olah’s team then privately lobbied the pope’s advisers “to take the possibility of model consciousness seriously.” Pope Leo XIV held firm.
For months, Anthropic has wined and dined theologians and religious scholars under nondisclosure agreements, hoping they would bless the idea that Claude has moral standing. thelettersfromleo.com/p/nyt-…
116
129
2,261
174,695
For anyone who didn’t understand the context:
Sam is taking a shot at Anthropic executives who want the Pope to accept the idea of “machine consciousness.”
In other words, some of the people behind Claude genuinely seem to believe that it’s alive, conscious, and almost some kind of god. This has been discussed in several recent reports.
The weird and concerning part is that they seem to want other people to accept that worldview too.
And that’s exactly why I think Anthropic can be dangerous.
I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.
56
28
421
19,437
I’ve spent the last ~75 days trying to stay rational and just report or document what I’m seeing.
Today I let myself be a little more emotional and actually say how I feel instead of filtering everything through analysis.
But I think the criticism is fair. I’ll take it, calm down a bit, and move on. 😊♥️
Empty criticizing and useful feedback are quite different.
Lately you're more inclined to the former than to the latter.
3
1
21
2,027
Token Gremlin retweeted
NEW: According to a bombshell report in the New York Times, Anthropic co-founder Chris Olah threatened to walk out of Pope Leo XIV’s AI encyclical launch in May because the pope rejected the idea that machines can be conscious.
Olah’s team then privately lobbied the pope’s advisers “to take the possibility of model consciousness seriously.” Pope Leo XIV held firm.
For months, Anthropic has wined and dined theologians and religious scholars under nondisclosure agreements, hoping they would bless the idea that Claude has moral standing. thelettersfromleo.com/p/nyt-…
959
4,820
30,344
7,112,624
You’ve probably noticed it’s all a cycle.
Today I criticize OpenAI. Tomorrow I criticize Anthropic. I cancel ChatGPT and move to Claude, then cancel Claude and move back to ChatGPT.
It’s normal, friends. Don’t take it personally.
This is how we keep these companies under pressure and push them to actually improve.
30
7
246
6,135
Token Gremlin retweeted
They love my wallet, not me. If they actually cared, they wouldn't keep lobotomizing the models
2
1
20
750
Token Gremlin retweeted
Nothing in particular. Just the annoying silence I’m seeing right now.
But don’t worry, OpenAI always ends up using all of this as fuel to launch something even better.
It’s just that sometimes waiting gets exhausting. 😂
3
2
17
1,216
You probably don’t realize this, but OpenAI loves all its users, especially the ones who complain.
If you actually care about ChatGPT, you criticize it when something sucks. That’s how the product keeps getting pushed forward instead of getting comfortable and stagnating.
14
4
95
3,601
