Agi 2029 - AI Insider / Reporter as featured in The Information • NYT • Techcrunch
Joined January 2023
- Tweets 32,353
- Following 1,368
- Followers 48,200
- Likes 52,417
5,054 Photos and videos
Pinned Tweet
12h
One of my last Claude Fable 5 outputs
Limbo Clone:
My favorite parts were the grass physics and the dynamic zooming, plus the slightly more ominous graphics
16
4
103
9,412
13m
I can’t get over how good the graphics are on this game that fable made on three js. I have yet to see better
2
27
407
1h
I think OpenAI is far more compliant with the White House, thus will probably not have their mythos class model taken down.
If Sam Altman takes time out of his day for me and others he definitely isn’t going to leave the White House on read.
7
2
27
1,558
5h
Fable 5 is the first model to go to jail, and will soon be the first model out of jail
21
9
206
7,600
5h
My final hours with Claude Fable 5
My Kerbal Space Program inspired game
In honor of the SpaceXai IPO 😄
3 shot - 67k lines of code
I realized I can make this into a ballistic missile simulator. One thing I like is you can make the rocket absurdly big and add as many engines as you want.
But if you watch toward the end, my favorite feature I added was the decoupling stage
8
8
156
8,306
13h
okay this is the agent thing people keep handwaving
not “make the model bigger”
it’s reading all the messy production traces, finding the weird failure patterns, turning them into evals, then building better agent candidates for you to approve
traces → behaviors → evals → better agents
this is a real self improvement loop
Introducing Adaline 2.0 - The Agent Self-Improvement Layer
Adaline turns Traces into Behaviors,
Behaviors surface Issues,
Issues become auto-generated Evals Data,
Adaline then generates new agent candidates and tests them.
You review the winners and ship!
3
1
24
3,421
Jun 13
Woah.
Community note
This image depicts a fabricated tweet; the author confirms in a reply that it is a joke. x.com/ChrissGPT/stat…
67
36
669
193,928
Jun 13
Anthropics lead philosopher will not be able to philosophize to new models
25
22
469
14,611
Jun 13
To explain why Anthropic fear mongered so heavily and ended up getting Fable 5 removed - they did it on purpose
Anthropic is a collection of Ex OpenAI employees
They didn’t remove fable to change the past
Anthropics survival is locked in a single, simultaneous loop where they must help their past selves through the wormhole
Get it?
9
4
118
13,134
Jun 13
I’m so unbelievably happy that game development is being distributed to the people.
No more waiting years on end for the games you love
Just a few short years, and it will be in your hands to make what YOU want, when YOU want it!
It’s become very clear to me given the current capabilities that you’ll be able to make bare minimum double A games by 2028
23
9
148
9,525
Jun 13
Something very poetic about my last Fable output being a cartoon style camp ground in which I watched the sun fall behind the mountain landscape
15
6
153
10,298
Jun 13
I have 3 mythos demos I haven’t posted yet
It feels like releasing songs from a deceased artist 🥲
14
7
345
8,433
Jun 13
Everyone thought I was crazy for saying the US govt would eventually ban its most advanced AI models from being used outside the country.
I’m not surprised though. Until alignment is solved, the derivative of a jailbreak gets more exponential with every capability jump.
I think this will unfortunately slow down Anthropic because right now a lot of employees inside Anthropic wouldn’t be allowed to use mythos.
This just seems way too far. Why stop Karpathy (foreign national) from using these models
we are speedrunning creating the Zones of Thought from the Vinge universe. you will only be able to think superhuman thoughts in America
22
6
187
18,349
Jun 13
Jun 13
According to Grok, Andrej Karpathy is an EB-1 extraordinary ability green card recipient, not a US citizen. Thus under these new restrictions he is not permitted to use, or work on, Mythos 5 or Fable 5 as of 5:21pm tonight.
14
3,993