@friedberg Isn't there a risk that this slows down progress and hence give an advantage to US's adversaries?
To minimize government control, what do you think about simply imposing liability and fines on labs whose models are used illegally?
Putting aside Anthropic's insincerity and manipulation, is throttling intelligence of ultra-capable models a winning approach? Doesn't it hold back AI adoption, give an opening for adversaries to overtake and win, and reinforce the negative public perception?
Feels like the government should find a different lever on the labs.
I would take an Opus 4.8-intelligence-level model that is 10x cheaper and faster over a Fable 5 any day, even if Fable is 10x more intelligent. GPT5.5 and Opus 4.8 are sufficiently capable for everyday development which is characterized by product and technical tradeoffs that require a human to steer. Sure, if you're running a huge technical migration of your entire codebase that is only bounded by passing tests then having a supermodel to get you through the finish line while navigating millions of tokens of context autonomously is great, but most everyday work is not that. Most everyday work is trading off product and architecture design decisions where there isn't a right or wrong. There's taste and judgement and iteration. That's where speed and cost matter more.
To xAI and Google - I think that's your opening to winning this crazy race. Probably requires several model architecture breakthroughs, but I'm rooting for ya!
For the same reason it is also great news for our comic creation app - our characters are more consistent across frames and you get more images per month. Check it out at https://t.co/VKI13IN1Is
✨ Nano Banana 2 is now live on 📸 Photo AI
It's a real breakthrough for image models because it FINALLY has high resemblance, or as Google calls it subject consistency
Which in my case means photos will actually look like you or your model trained, not "somewhat" like you, no really you, which is the entire point of Photo AI from the start
Also they made Nano Banana 2 about ~2x cheaper vs Nano Banana Pro, which helps!
I was spending $40,000/mo (like 40% of revenue now) on it so I will save ~$20,000/mo now, which I think I need to to get better profit margins (I like to stay at about 80-90% profit)
אנתרופיק יודעים לעשות מודלי AI אבל לא יודעים לעשות שיווק. מהלך מטומטם מצידם לדרוש את שינוי השם של המוצר. היו ��ריכים לרכב על הטרנד כדי לבסס את המודל שלהם כמודל המוביל למוצרי לולאת-אייגנט. אם הייתי סם אלטמן, הייתי קונה היום את Clawdbot ומשנה את השם לצאטבוט או משהו בסגנון