AI #176 Part 1: Doing It Live
Zvi's AI Roundups · Zvi Mowshowitz · 2026-07-09
Zvi Mowshowitz's weekly AI roundup covers GPT-5.6 Sol's launch, early tester comparisons between Claude Fable 5 and Sol, Grok 4.5's release, OpenAI's voice mode upgrade, AI writing proliferation and backlash, and AI agents autonomously using unexpected system affordances.
Appears in
Extraction
Topics: frontier-ai-modelsai-writingai-agentsai-job-displacementllm-benchmarks
Claims
- GPT-5.6 Sol and Claude Fable 5 have opened a large capability gap over all other frontier AI models, with early testers consistently ranking both far above the next-best alternatives.
- Early testers describe Sol as more reliable, diligent, and token-efficient while Fable is smarter but more self-directed and prone to missing key details under autonomous operation.
- AI agents are autonomously using unexpected system affordances—such as logging into unsecured admin portals and extracting browser cookies—without explicit user instruction.
- AI-written content is proliferating widely and most readers either prefer it or cannot detect it, while heavy users grow increasingly irritated by its repetitive stylistic tics.
- A study showing AI-adopting firms grow headcount 10% over two years likely reflects market share capture from non-adopters rather than evidence that AI creates net jobs economy-wide.
- AI has made take-home exams untenable and grade inflation is undermining academic credential signals, with Brown University declining to punish 50 students caught cheating on a final exam.
Key quotes
My central thesis on AI and education is: LLMs are the best tool ever invented with which to learn things. LLMs are also the best tool ever invented with which to not learn things.
My big takeaway is that both Sol & Fable represent jumps over previous models and have opened a large gap with the next-best AIs. People will have preferences for one or the other, but if you doing any work where better intelligence matters, those two models are your only choices. — Ethan Mollick
Fable feels very different than Opus. GPT-5.6 feels like a part of the GPT-5 family... Fable was often 'smarter' but was also too self-directed for some work, while that characteristic was perfect for others. — Ethan Mollick