Models, synthetic data, and environments are a flywheel now. The best model generates the best data, trains a better model, generates even better data. Rinse and repeat. But the flywheel only spins where you can verify: math, code, agents with graders. Everywhere else it merely
LLMS
-
JSON Extras API Explorer with Fable 5 and GPT-5.5
By
–
Since the JSON extras API is a bit difficult to explain without an example, I also had Fable 5 and GPT-5.5 collaborate on this custom API exploration tool to test the new feature https://tools.simonwillison.net/datasette-extras-explorer#url=httpslatest.datasette.iofixturesfacetable.json_size1_facetstate&extras=countcount_sqlexpandable_columnsfacet_results
… -
SambaCloud runs Gemma 4 31B 30% faster than others
By
–
Gemma 4 31B is one of the strongest open models available today. And it's running fastest on SambaCloud. According to @ArtificialAnlys
, SambaCloud delivers Gemma 4 31B over 30% faster than the next provider. Why devs care: -89.2% AIME 2026
-80.0% LiveCodeBench v6
-Native -
GPT-2 fears reminiscent of boy who cried wolf
By
–
they were terrified of GPT-2. definitely into boy who cried wolf territory.
-
Fable’s AI attempt to complete Coleridge’s Kublai Khan
By
–
Fable's attempt to complete Kublai Khan. Better, though no Coleridge: https://
claude.ai/public/artifac
ts/d7d3351f-5ad5-4d73-a644-4a1426abe558
… The most interesting thing is that it thought for 10 minutes & the thinking trace is full of pretty complicated (seeming?) musings about Coleridge's intent. A little literal, though. -
Stop doing things that don’t matter, use MCP to dub to Mandarin
By
–
— Pika (@pika_labs) 11 juin 2026
Stop doing things that don't matter. Instead do things that resonates with yourself I love making communication easy, and tearing down barriers using technology Here I used Pika's MCP to quickly dub myself to mandarin, because I like this stuff Time for me to go global x.com/pika_labs/stat…
-
Rippling’s AI-native transformation across all products in 6 months
By
–
How they shipped this in 6 months https://
langchain.com/blog/how-rippl
ing-went-ai-native-across-every-product-in-6-months-with-deep-agents-and-langsmith
… -
Pew: One in five US teens get news from AI chatbots
By
–
6/ Here is who is doing it. Pew found one in five US teens now get their news from ChatGPT, Claude, or Gemini. The people still forming their instincts about truth are handing those instincts to a model.
-
Anthropic fears Mythos misuse, fails to explain safeguards
By
–
Two things are true:
(1) Anthropic (or parts of it) are absolutely and sincerely worried about the misuse of Mythos-class models & have put in excessive safeguards until they are confident it will not be misused
(2) They have not succeeded in explaining/convincing people of this -
Speculative decoding makes LLMs 8.5x faster without accuracy loss
By
–
Researchers found a way to make LLMs 8.5x faster!
— Akshay 🚀 (@akshay_pachaar) 11 juin 2026
(without compromising accuracy)
Speculative decoding is quite an effective way to address the single-token bottleneck in traditional LLM inference.
A small "draft" model first generates the next several tokens, then the large… https://t.co/JCdqjCKcKU pic.twitter.com/HbKmRqdF5PResearchers found a way to make LLMs 8.5x faster! (without compromising accuracy) Speculative decoding is quite an effective way to address the single-token bottleneck in traditional LLM inference. A small "draft" model first generates the next several tokens, then the large