The key distinction: BPE gives us a strong procedure. ConvexTok gives us a procedure, an optimization relaxation, and a certificate. That moves tokenisation from engineering folklore toward measurable optimality.
RESEARCH
-
Paper: Tokenisation via Convex Relaxations reframes tokenizers
By
–
The tokenizer is an architectural prior disguised as preprocessing. And almost everyone has been treating it like plumbing. A new paper by Jan Tempus, Philip Whittington, Craig W. Schmidt, Dennis Komm, and Tiago Pimentel changes the frame: Tokenisation via Convex Relaxations
-
Tokenization as Architectural Prior: New Research Paper
By
–
The tokenizer is an architectural prior disguised as preprocessing. And almost everyone has been treating it like plumbing. A new paper by Jan Tempus, Philip Whittington, Craig W. Schmidt, Dennis Komm, and Tiago Pimentel changes the frame: Tokenisation via Convex Relaxations
-
AI Solves 1946 Math Problem Using Ordinary System
By
–
🤖 AI cracked a math problem that stood untouched since 1946 and the scary part is how ordinary the system was.#mathmatics #artificialintelligence #AI #maths @AlbertoEMachado @Eli_Krumova @postoff25 @Khulood_Almani @anand_narang @NutritiousMind @baski_LA @TanyaSinha_ @devaang… pic.twitter.com/xlZLAlqdIs
— Amitav Bhattacharjee (@bamitav) 23 mai 2026AI cracked a math problem that stood untouched since 1946 and the scary part is how ordinary the system was. #mathmatics #artificialintelligence #AI #maths @AlbertoEMachado @Eli_Krumova @postoff25 @Khulood_Almani @anand_narang @NutritiousMind @baski_LA @TanyaSinha_ @devaang
-
Anthropic’s Powerful Cybersecurity AI Model Shared Selectively After Revealing 10,000+ Vulnerabilities
By
–
Anthropic created a cybersecurity model so dangerous that it decided not to release it to the public. Instead, it gave it to companies like Apple, Google, Microsoft, AWS, and Cloudflare. And in just one month:
→ 10,000+ vulnerabilities found
→ bugs hidden for 27 years -
AI Progression: AGI, ASI, and Convergence Stages
By
–
C'est facile de répondre à cette question. Nous avons 3 étapes à venir après l'IA. Prochaine étape, AGI.
Puis à l'étape suivante, ASI.
Et pour finir, CBT. AGI = Artificial General Intelligence (human-level)
ASI = Artificial Superintelligence (human surpassed)
CBT = Convergence -
Montreal.AI launches public forum for AGI-first to ASI-first era
By
–
MONTRÉAL.IA / http://
MONTREAL.AI is live on Eventbrite. Public Intelligence for the AGI-first → ASI-first Era. Briefings. Debates. Archives. Public record. http://
MONTREAL.AI convenes the public-intelligence forum for frontier AI, sovereign intelligence, -
Frontier Models and Scientific Progress Forecasting Study
By
–
Can frontier models forecast scientific progress? Mostly no, but here is why. This work looks at 4,760 scientific events across disciplines. Frontier models can identify plausible research directions when given options. They cannot reliably predict whether an advance will land,
-
Lightning Indexer DSA Top-k Inference Path Demo
By
–
It’s just demoing the inference-time DSA top-k selection path at this point. The Lightning Indexer has trainable parameters, but this version does not train them (yet)
-
Gemini 3.5 Flash Achieves Pareto Frontier on Vending Bench
By
–
Gemini 3.5 Flash is on the Pareto frontier of cost per intelligence on Vending Bench (a measure of a models ability to run a simulated store)!