At the minimum, transparency in what queries lead to which outputs (interpretability), how this connects to educational principles/pedagogy, more evidence of how this may impact cognition, especially of developing brains, in the short, medium and long term.
RESEARCH
-
RLHF requires pairwise comparison, not thumbs-up/down
By
–
The thumbs-up/down buttons aren’t for RLHF. The HF in RLHF is selecting the better of two outputs for the same prompt (or, if you have more than two, ordering them best-to-worst). Pointwise data (thumbs, star ratings, etc.) generally doesn’t work well AFAIK.
-
Grok 4.3: 10x cheaper, close to frontier models
By
–
Grok 4.3 is literally 10x cheaper than GPT-5.5 or Claude for token output costs. It's also shockingly close on benchmarks (or better in some long-horizon agent tests) to the frontier models.
-
Opus outlines, GPT-5.5 drafts and executes
By
–
This works beyond coding. Content strategy: use Opus to outline the argument structure. Use GPT-5.5 to draft. Research: use Opus to design the research methodology. Use GPT-5.5 to execute the analysis. Business planning: use Opus to define the framework. Use GPT-5.5 to
-
Opus 4.7 plus GPT-5.5 nearly doubles benchmark scores
By
–
Dan Shipper at Every tested this on their Senior Engineer Benchmark. The scores:
Opus 4.7 alone: low 30s GPT-5.5 alone: low-to-mid 40s Opus 4.7 planning + GPT-5.5 executing: 62.5 For reference, human senior engineers score 80-90. The combo nearly doubled either model's -
Jack Clark Predicts Fully Automated AI R&D Timeline 2027-2028
By
–
Fully automated AI R&D: ~30% chance by the end of 2027, ~60%+ chance by the end of 2028 Overall, Anthropic's Jack Clark has written a very worthwhile essay: His timeline is that fully automated AI R&D probably won’t arrive in 2026, but we may see a proof-of-concept within 1–2
-
Why 1M-Context Models Still Don’t Work Beyond 200K Tokens
By
–
it is endlessly fascinating to me that we still don't have a true 1M-context model it's an unusual case where the infra is far ahead of the science. Claude discontinued 1M+ context bc it didn't really work past ~200k we don't have the right data? training techniques? not sure
-
Turing Test Conflates Intelligence With Human Behavior
By
–
The Turing test is a test of intelligence. It has nothing to do with consciousness. (And even as the former it’s a bad idea, because it confuses being intelligent with being human. It’s like saying an airplane doesn’t fly unless it can fool birds into thinking it’s a bird.)
-
GPT-4o and Llama 3.3 Pose Low Harm Risk Study Suggests
By
–
I think the fact that GPT-4o and Llama 3.3-80B did no significant harm is just as important as whether AI helped. If older (less accurate & more sycophantic) chatbots essentially did nothing for people who followed their advice, it means that there is less risk of harm as well.
-
Stanford’s Half Billion Dollar Machine: Creating Things Through Education
By
–
I was the first to show you Levangie about a year ago. And is how I build: https://t.co/kiuZ7QXLzb which watches the whole AI community here. Congrats @blevlabs for stepping up your business. The Residency rocks. https://t.co/PGCfr6JTeo
— Robert Scoble (@Scobleizer) 4 mai 2026I was the first to show you Levangie about a year ago. And is how I build: https://
alignednews.com/ai which watches the whole AI community here. Congrats @blevlabs for stepping up your business. The Residency rocks.