MaxxBench adds 89 DeepSWE tasks with small instruction or code changes, testing model consistency. Models around 74% accuracy vary by 23 percentage points across tasks.
Key points
MaxxBench includes 89 DeepSWE tasks with small changes to the instructions or starting code.
Models clustered around 74% on DeepSWE end up 23 pp apart.
Sources
AI 3D cartoon model shape similarity 94/100 score
AI models built 3D cartoon characters and were compared to a professional artist's model. The best AI model achieved a 94/100 score for shape similarity but got the facial expression wrong, appearing wide-eyed versus the artist's serious look.
Key points
Shape similarity score: 94/100 for the best AI model.
Facial expression difference: AI model looks wide-eyed, artist's model looks serious.
Sources
Claude GTM research method reduces 4 weeks to 3 hours
Claude GTM research method claims to turn 4 weeks of market research into 3 hours by feeding raw material and asking forced-thinking questions. It outlines data sources, question prompts, and a follow-up critique step, then suggests using Gojiberry to test hypotheses with buyers.
Key points
Raw material: 25-40 competitor websites, recent earnings calls from 3-5 biggest incumbents, 100+ customer reviews, and every Reddit or forum thread about the category.
Questions: “What do the winners in this market know that their customers would never admit?” and “List the beliefs this whole industry takes for granted. For each one, what would need to happen for it to be false?”
Follow-up: Ask Claude to “Steelman this thesis like a top-tier VC would. Then show me exactly where it fails.”
Testing: Use Gojiberry to find and reach the right people for buyer conversations.
Sources
Starship Flight 15 launch targeting November 13
Starship Flight 15 is targeting a launch as early as Friday, November 13, pending regulatory approval. The post expresses confidence it will succeed.
Key points
Flight 15 targeting launch as early as Friday, November 13
Launch pending regulatory approval
Sources
Manufacturing jobs jump to highest since Dec '24
Manufacturing jobs jumped in Oct to the highest level since Dec '24. The sector had a 3-year slide before a turnaround began this year, after hemorrhaging jobs at the start of ’23.
Key points
Jobs jumped in Oct to highest level since Dec '24
Sector was in a 3-yr slide before turnaround started this year
Sources
Compaction improves, 3-tier subagent architecture recommended
Compaction is much better now so you may not need to explicitly clear context for model focus. Autocompact is usually set to 400k and manual compact runs between 200-400k. A three-tier subagent architecture is suggested, with no need for more than three nesting levels for most tasks.
Key points
Autocompact typically set to 400k tokens
Manual compact performed between 200-400k tokens depending on model/harness
Top-level orchestrator can launch subagents that launch further subagents
No good reason to exceed three levels of subagent nesting for most tasks