The caveat is the important bit: performance drops on raw, uncurated USPTO.
Takeaway: reaction completion is not a side quest. It is part of the foundation for trustworthy reaction prediction, retrosynthesis, and process intelligence. 5/5 https://t.co/XfVzZK46ik
Reaction prediction models are only as good as the reaction data they learn from.
This paper is worth a look: CompleteRXN: Toward Completing Open Chemical Reaction Databases 1/5
Their Constrained Reaction Balancer adds atom-balance constraints during decoding. It reaches 99.20% equivalence accuracy on random splits and 91.12% on extreme OOD splits. 4/5
5/ Token billing itself is transitional. Salesforce is already moving to "agentic work units" — paying for outcomes, not inputs.
Three pricing eras:
Flat-fee (2022–26): dead
Per-token (2026–~27): now
Per-outcome (2027–30?): coming
Full breakdown ↓ https://t.co/VsBCunzHKW
1/ Uber burned its entire 2026 AI budget by April. ServiceNow did the same. KPMG says U.S. orgs project $207M average AI spend over the next 12 months — nearly 2x last year.
The all-you-can-eat era of AI coding tools is over. The deadline matters.
4/ The part nobody is talking about: 2 cost drivers are 100% within enterprise engineering control.
① Context bloat — sending the whole codebase on every call
② Routing every task to Opus when Haiku would do (5x cheaper)
Estimated savings: ~$9.6M/yr for a 500-engineer org.
SkillOpt is a strong signal that agent “skills” should stop being static docs.
The paper treats a best_skill.md as a trainable adaptation layer: agents run tasks, traces are reviewed, bounded skill edits are proposed, and only validation-improving edits survive.
That gives you:
- eval-backed skill evolution
- compact auditable artifacts
- rejected edits as negative feedback
- no model weight updates
- transferable procedural knowledge
It feels closely related to Karpathy’s autoresearch, but with a safer target: optimize the agent’s procedural skill before letting it mutate core code.
My take: a skill without evals is just advice. Useful, but fragile.
#AIAgents #LLM #AgenticAI #MLOps
HOLY CRAP Trump actually accomplished a miracle. Here is what he got out of Iran:
- Reduce its stockpile of enriched uranium by about 98%
- Limit uranium enrichment to 3.67% purity (far below weapons-grade)
- Cut the number of installed centrifuges by roughly two-thirds
- Only enrich uranium at one declared site (Natanz)
- Stop enrichment activities at Fordow and convert it into a research facility
- Redesign the Arak heavy-water reactor so it could not easily produce weapons-grade plutonium
- Ship out or dilute excess enriched uranium
Allow extensive inspections by the International Atomic Energy Agency (IAEA)
Permit continuous monitoring of nuclear facilities and supply chains
- Accept “snap” inspections under expanded monitoring rules
- Avoid building new heavy-water reactors for years
- Stay within strict limits on uranium stockpile size and centrifuge development for set periods ranging from 10–25 years
Ooops, sorry!
That was the JCPOA that Obama signed with Iran, only to have him tear it up, kill 140 kids, get hundreds of Americans injured, 13 killed, and gas prices to surge 50%.
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
Missouri just voted out every council member who approved a $6B AI data center. All four. Gone.
The tech industry called it NIMBYs killing $1 trillion in value. But 70% of Americans now oppose data centers locally — higher than peak nuclear opposition. Ever.
The problem isn't irrational fear. It's that electricity bills near data center clusters have surged up to 267% in five years (Bloomberg). Communities bear the cost. Shareholders capture the benefit. Nobody fixed the deal.
Loudoun County, VA hosts the world's densest data center market. Schools funded, housing built, property taxes down. Thirty years of minimal opposition. The model exists.
The industry just keeps choosing Festus over Loudoun.
My full article in the replies
RAG might already be becoming obsolete.
A month ago, Andrej Karpathy dropped a simple GitHub gist called “LLM Wiki.”
Now the comments section looks like the birth of an entirely new AI category.
5000+ stars later, developers are rapidly building:
• persistent AI memory systems
• self-maintaining knowledge bases
• multi-agent research environments
• contradiction detection engines
• AI-native company operating systems
• local-first memory architectures
• graph-based reasoning layers
• evolving second brains
And the craziest part?
Most of them were built in DAYS.
Because the core idea is insanely powerful:
Instead of AI repeatedly retrieving raw chunks like traditional RAG…
…the model continuously maintains a living knowledge system.
Not temporary context.
Persistent synthesis.
The shift sounds subtle until you realize what it changes:
RAG:
retrieve → answer → forget
LLM Wiki:
ingest → synthesize → evolve
That one architectural difference is causing an explosion of experimentation right now.
People are already building:
• agent memory operating systems
• AI-maintained engineering documentation
• self-healing knowledge graphs
• persistent research environments
• conversational memory architectures
• contradiction-aware wikis
• context compression engines
• machine-readable company systems
The comments section alone feels like watching an ecosystem form in real time.
One developer built deterministic contradiction detection using sheaf cohomology
Another built “sleep consolidation” for AI memory systems inspired by human memory formation
Another created persistent multi-agent vault conversations
Another turned entire repositories into continuously maintained AI wikis
Another built local-first memory systems with audit trails, provenance, graph exports, and MCP integration
This is the important part:
Karpathy didn’t launch a product.
He introduced a pattern.
And patterns are what create ecosystems.
The same way:
• transformers created modern AI
• RAG created AI retrieval startups
• agents created orchestration frameworks
LLM Wikis may create persistent AI memory infrastructure.
That’s why this moment feels different.
For years, AI systems have been stateless.
Now developers are trying to build systems that actually accumulate understanding over time.
And once knowledge compounds instead of resetting…
…the entire interface layer of AI changes.
(Link in comments)