Tomorrow’s AI News video is a big one for multiple reasons (including a fun announcement). Here are all the highlights from the past 2 weeks, what did I miss? – @OpenAI launched ChatGPT Pulse
– OpenAI unveiled age prediction + parental controls
– @LumaLabsAI released Ray3
–
LLMS
-
OpenAI ChatGPT Pulse and Luma Ray3: Major AI Launches
By
–
-

Flexible parsing unlocks hidden capabilities in weaker language models
By
–
Flexible parsing revealed hidden capabilities in weaker models LLM-based parsers captured valid answers from free-form reasoning, improving reported accuracy for models that struggled with rigid constraints. It’s a trade-off: exactness versus resilience.
-
Parsing and Evaluation: Technical Foundations for AI Systems
By
–
The bottom line: Parsing isn't just a technical detail. It's part of the evaluation story. Don’t forget to follow us @snorkelai and @realjustinbauer – and if you have questions about evals, RL environments, and expert-developed datasets, talk to us!
-

Structured Formats Limit AI Reasoning Capabilities
By
–
Structured formats can actually constrain reasoning Models like GPT-4.1 and Grok-3 performed worse when forced into rigid JSON structures. The formatting requirements limited their ability to think through complex problems.
-
Reasoning Models Show Resilience Across Parsing Methods
By
–
Reasoning-first models stayed resilient Claude Sonnet 4, Gemini 2.5 Pro, and o4-mini showed minimal sensitivity to parsing methods—their strong reasoning held steady across formats.
-
OpenAI Models Solve Programming Challenges With Code Sandbox
By
–
Confirmation from Ahmed El-Kishky – who worked on this project at OpenAI – that their models solved the programming challenges with a code execution sandbox but no internet access
-
Taking Extra Care with Model Decoding and Out-of-Distribution Constraints
By
–
okay maybe the question is – how would one take “extra care”? the D is the D, cant be touched. and yet people are constrained decoding thousands of tokens with presumably only slight intelligence drop. maybe my json example is too light of a constraint. the OODness would
-
Models Writing Tool Calls in Code Is More Expressive Than JSON
By
–
That's a good direction! Let models write their tool calls in code snippets, it's much more flexible & readable & expressive than the lame JSON blobs of usual tool calling!
-

OpenAI plans 125x energy capacity increase by 2033
By
–
OpenAI have 9x'd their capacity this year, and have a goal to 125x further by 2033, thus exceeding current energy capacity of the whole of India – as per @alexeheath reporting of @sama
's internal slack note. One thing to note is that 125x energy capacity increase understates the -
Realtime Voice AI Still Too Limited for Meaningful Debate
By
–
Disappointed by the lack of progress on the realtime voice AI front. They'd be so good for debating ideas if they weren't as dumb as rocks.