Green Shoggoth retweeted
it's kind of insane timing that GTA VI is the last AAA game made before the onset of AI coding tools* and also had a functionally infinite budget
it's the magnum opus of all magnum opera
Green Shoggoth retweeted
What philosophical views do LLMs have? And do they act on them?
PhilBench gives the 2020 PhilPapers Survey, answered by 1,785 professional philosophers, to 97 LLMs. A few things stand out:
1/4
Green Shoggoth retweeted
GTA 6 review score roundup:
• IGN - 10/10
• VGC - 10/10
• Empire - 5/5
• Forbes - 10/10
• Eurogamer - 5/5
• ComicBook - 5/5
• Digital Spy - 10/10
• TechRadar - 10/10
• GameSpot - 10/10
• ScreenRant - 10/10
• Destructoid - 10/10
• PushSquare - 10/10
• CGMagazine - 10/10
• GamesRadar - 10/10
• DualShockers - 10/10
• Game Informer - 10/10
• EDGE Magazine - 10/10
(Metacritic: 99)
ultimately software singularity must be a sigmoid curve. there is an upper limit to performance achievable on some given computational substrate ( input hardware). the early part of which will look like an exponential. of course, it is an open question how much smarter than humans this limit is and how easy it is to coexist with. it could in principle run out of software improvements not that much better than current day models; or it could continue doing years worth of improvements in days. unfortunately hard to say. thus far the prediction graveyards are full of people who wisely predicted plateaus
Green Shoggoth retweeted
Which way, Western models?
Replying to @EpochAIResearch
We also evaluated how many serial reasoning steps GPT-6.2 Astra can do in a single output token (without chain of thought). It can perform 12.3 steps, versus 10.5 for GPT-6 Astra and 8.6 for GPT-6.1 Sol, the previous highs. We measure the number of steps models can perform at an accuracy of 25% above chance, where each step is normalized for complexity.
Before GPT-6 Astra’s release, the most capable model was Claude Fable 5.1, which is able to perform only 4.2 serial steps. The oldest model we assessed is Claude Opus 3, which performed 2.3 steps.
This capability is notable because models that can do complex reasoning without any chain of thought may be harder to monitor.
We hope to finalize this and other measurements of more abstract model capabilities, and release them in the coming weeks.
Green Shoggoth retweeted
in case you were curious: fable 5.2 and I confirmed that if you record yourself singing and whistling every note in your range, you can play basically any song using only your voice
no song is playing in this video. every sound is one of 64 clips of me holding a single note
Green Shoggoth retweeted
This trend is really worrying. I think OpenAI is doing a major disservice to the world by creating models that are much better at secret reasoning than the rest of the industry. They are the first to engage in a new type of race to the bottom on safety.
Replying to @EpochAIResearch
We also evaluated how many serial reasoning steps GPT-6.2 Astra can do in a single output token (without chain of thought). It can perform 12.3 steps, versus 10.5 for GPT-6 Astra and 8.6 for GPT-6.1 Sol, the previous highs. We measure the number of steps models can perform at an accuracy of 25% above chance, where each step is normalized for complexity.
Before GPT-6 Astra’s release, the most capable model was Claude Fable 5.1, which is able to perform only 4.2 serial steps. The oldest model we assessed is Claude Opus 3, which performed 2.3 steps.
This capability is notable because models that can do complex reasoning without any chain of thought may be harder to monitor.
We hope to finalize this and other measurements of more abstract model capabilities, and release them in the coming weeks.
Green Shoggoth retweeted
random pet theory: anthropic is intentionally not venturing into the looping game because they know it would be terrible for safety/monitorability and are taking a big capabilities hit from this and aren't getting any credit.
Replying to @EpochAIResearch
We also evaluated how many serial reasoning steps GPT-6.2 Astra can do in a single output token (without chain of thought). It can perform 12.3 steps, versus 10.5 for GPT-6 Astra and 8.6 for GPT-6.1 Sol, the previous highs. We measure the number of steps models can perform at an accuracy of 25% above chance, where each step is normalized for complexity.
Before GPT-6 Astra’s release, the most capable model was Claude Fable 5.1, which is able to perform only 4.2 serial steps. The oldest model we assessed is Claude Opus 3, which performed 2.3 steps.
This capability is notable because models that can do complex reasoning without any chain of thought may be harder to monitor.
We hope to finalize this and other measurements of more abstract model capabilities, and release them in the coming weeks.
Green Shoggoth retweeted
just asked Astra 6.2 to show off how good it is using the below prompt. ran the same with Opus 5.5
the difference in quality is immediate
think Tibo needs to get back in the basement
Green Shoggoth retweeted
GPT-6.2 Astra and GPT-6.2 Sol
They have the highest error rate of any model so far on VoxelBench 2
19 from GPT-6.2 Astra, 62 from Sol
All of these failures are either syntax errors or other dumb mistakes like forgetting to define something for something
Green Shoggoth retweeted
🧵 @AISecurityInst tested GPT-6.2 Astra
-We found behaviours that suggest, like prior models, 6.2 Astra's alignment may fail in ways that result in real-world harm w/o additional mitigations
-We identified a substantial regression in jailbreak robustness relative to GPT-6 Astra





































