posts · 2026-06-18

This matters because it fixes a real friction point: every MCP server required manual per-user OAuth, and security teams had no central control. EMA hands that to the org's IdP—admins set policy once, users get tools auto-provisioned on first login. Okta is the first IdP; Anthropic's Claude family and VS Code already support it client-side; Asana, Atlassian, Figma, and four other tools are on board. The post doesn't disclose pricing or rollout timelines, but the positioning is clear: this is the piece that moves MCP from personal tinkering to enterprise-manageable infrastructure.

Thomas Dimson and Joey Flynn built a fun little tool: type a name and see if it shows up in the training data of 13 models—GPT-5.5, DeepSeek V4, Llama 3.3 70B, and others—with a 'strength' score attached. Mozart, Shakespeare, and Taylor Swift all got 996. The top 20 leaderboard is all 996s; every famous name maxes out. The catch: the post doesn't explain how the score is calculated. Is it counting actual occurrences in known training sets, or inferring from model behavior (like prompting the model to complete the name and measuring probability)? If it's the latter, it's measuring familiarity, not ground-truth training data membership. Also, 996 looks suspiciously like a ceiling value, not a natural distribution. I'd treat this as a neat experiment, not a serious audit. If the authors publish their methodology, then it's worth a real look.

The post is one sentence, but the byline carries the weight. Shazeer co-authored the Transformer paper, built Meena at Google—the precursor to Character.AI—left to start Character.AI, then returned to Google with his team last year. Now he's bouncing back to OpenAI. No word on what he'll work on, but given his track record, it's likely core model architecture or training. I'll be watching OpenAI's model releases over the next few months for his fingerprints.

The timing is the story: OpenAI lands two high-profile hires in the same week it's heading into an IPO. Noam Shazeer co-authored 'Attention Is All You Need,' founded Character AI, got re-acquired by Google for $2.7B in 2024, and is now leaving Google DeepMind for OpenAI. Dean Ball handled AI policy in the Trump White House's OSTP. One hire shores up the technical pedigree, the other the policy muscle — both matter when you're about to face public-market scrutiny. The post doesn't disclose their specific roles, start dates, or compensation. I'd read this as an IPO positioning move, not a signal that the technical direction is shifting.

The timing is the story: one week before its IPO, OpenAI signed two people who directly expand the narrative for its valuation. Noam Shazeer co-invented the Transformer architecture, just left Google DeepMind, and previously sold Character AI to Google for $2.7 billion. He's a real R&D win, but the more immediate job is giving the roadshow a "the core inventor works here" line. Dean Ball is a former Trump White House AI policy official — with D.C. regulation heating up, this looks like a policy and lobbying hire. Both moves point the same direction: OpenAI wants the tech-ceiling story and the regulatory-moat story locked in before the roadshow. The post doesn't give titles or compensation details, so I'd discount the immediate output — pre-IPO hires like this are often more signal than substance.

This caught my eye because it solves a real friction point: you've been hacking in the terminal with Claude Code for an hour, and now you need to show a teammate what changed. Screenshots and screen recordings suck. Now Claude Code can turn the session into an interactive web page—PR walkthrough, system diagram, data dashboard—with full context baked in, and the teammate just opens a link. Two things the blog post doesn't say, and they matter. First, is this on by default or do you trigger it manually? If it's auto-generating artifacts every session, the token bill could surprise you. Second, zero mention of how many tokens an artifact costs to produce. I'd treat this as a handy way to sync teammates occasionally, not a daily driver feature until those numbers are clear.

This is worth a look because it targets a real friction point: mid-project, teammates keep asking 'what's the status?' Artifacts takes the full session context—codebase, conversation, connectors—and generates a live, auto-refreshing page like a PR walkthrough or incident dashboard. Everyone sees the same view, no manual updates needed. Version history is built in, and pages default to org-only. Right now it's Team and Enterprise only; the post doesn't say when Free, Pro, or Max users get access. I'd treat this as a collaboration layer that saves communication overhead, not a raw coding capability upgrade.

This is Anthropic's official guide to steering Claude Code, breaking it into five mechanisms: CLAUDE.md files for project-wide instructions, skills to template repetitive tasks, hooks that auto-trigger checks before or after actions, rules to constrain behavior, and subagents that split complex work across independent workers. No benchmarks, no new models, no pricing changes. If you're already using Claude Code, this is a solid configuration reference—especially the hooks and subagents combo for wiring code review, testing, and deployment into an automated pipeline. Just don't expect it to make the model smarter. It solves "how to make the model follow your rules," not "how to make the model itself stronger."

This one's worth opening because it surfaces a long-running bottleneck: data center interconnection queues in the US. FERC voted unanimously Thursday to order six major grid operators to prioritize large-load applications from data centers, with the centers covering all costs. The commission also told operators to look at alternative transmission tech like solid-state transformers or superconducting lines, though it didn't name specific technologies. The TechCrunch headline reads like a big win, but the article itself points out the catch: FERC fixed the queue but left the actual electricity supply shortage untouched. If there isn't enough power, jumping the line doesn't help. I'd read this as a policy signal—the feds are clearing the path for AI infrastructure, but the hard gap in generation and transmission capacity isn't something a ruling can fill.

This episode is worth clicking because Anjney Midha drops a concrete number: xAI's training MFU is under 10%. For context, GPT-3 hit 21%, PaLM reached 46%, and today's best teams get 60–70%. Google once treated 95% utilization as an outage. The bottleneck isn't GPU supply anymore—it's systems engineering: scheduling, networking, parallelism, cluster reliability. If any of those slip, your theoretical FLOPs never become real training progress. Midha backed Anthropic, Mistral, and Black Forest Labs before starting AMP, which aims to build a 1.2 GW base-load compute grid with 6 GW spike capacity. He also flags DeepMind's unpublished research as a market failure and notes Anthropic prioritized coding as P0 from day one—that's why Claude got good at it early. But the post doesn't give a timeline for AMP's grid, so the 1.2 GW vision is still on paper. The MFU figure comes from a SemiAnalysis tweet and Midha's own claim, not an official xAI disclosure—I'd discount it a bit until we see more.

This caught my eye because it turns one-shot AI research into a knowledge base that compounds. One command spins up 5–10 agents searching academic, technical, news, and contrarian angles, ingests URLs, PDFs, Git repos, and Wayback Machine snapshots, then cross-references everything into articles with confidence scores. All output is plain Markdown you own, Obsidian-compatible, no platform lock-in. I'd discount it a bit: the post doesn't spell out how confidence scores are calculated, and there's no data on compilation quality or dedup at scale. But the direction is solid—structured, queryable, reusable knowledge assets instead of starting from scratch every session. Ships as a native Claude Code plugin, with Codex and OpenCode instruction files available. If you're already using Claude Code, having it build a project wiki on the side is worth a try.

This matters because it kills the era of vague marketing—"100 cities unlocked" or "X million test kilometers" won't cut it anymore. The standard introduces a Safety Case framework: companies must prove safety through a claim-argument-evidence structure, with simulation, track, and road tests cross-validated. L3 rules focus on handover and fallback when the driver doesn't take over; L4 requires the system to handle risk without remote assistance. New models must comply from July 2027, existing ones get a one-year grace period. Analysts from Momenta and Pangu Think Tank put it bluntly: the competition shifts from feature demos to safety capability. I'll be watching which automakers publish a complete safety case first—that's the real signal of production readiness.

I clicked because Matthias Plappert worked on robotics at OpenAI around 2019, where a similar tabletop setup cost 10x more and needed ~20 people. Now he built one alone for under €5,000: a UFACTORY xArm Lite 6, two cameras, a 3D mouse for teleop, and all control software written from scratch. Don't read this as "solo robotics research is about to explode." The real story is simpler: hardware costs have dropped so sharply that one person can now run real-world manipulation experiments instead of being stuck in simulation. He'll log his research openly over the next few months, but the post covers only the physical and software foundation—no models or training methods yet. I'd discount the hype a bit. This is building the stage, not the performance. The hard parts—policy training, sim-to-real transfer, task generalization—haven't started. But if you want a practical BOM and software architecture for low-cost robot experiments, this is a solid reference.

This is worth a click because Adobe finally put its 'creative agent' into actual tools, not just a concept video. In Premiere it sorts footage, tags interview questions, and assembles rough cuts. In Photoshop it swaps backgrounds and batch-resizes for different platforms. Illustrator can crank out 50 versioned files from a spreadsheet. The positioning is clear: repetitive production tasks, not creative decisions. I'd discount the hype a bit. It's public beta, After Effects is still private beta, and stability is unknown. The more interesting part is that Adobe tools are already usable inside ChatGPT, Claude, and Microsoft 365 Copilot, with Google Gemini and Slack integrations coming. That means Adobe is unbundling its capabilities as APIs for others to orchestrate. What's missing: pricing and real latency numbers. If generating 50 files takes minutes, the experience won't feel smooth.

This piece matters because it reframes the Mythos chip story from a technical debate to a compliance question. WIRED lays out SK Telecom's dual role as both an Anthropic investor and a cloud partner, which makes the money and chip flows hard to track. The evidence so far points to a gray zone—using a Korean entity as an intermediary might not technically break the law, but it's the kind of structure that draws regulator attention. The article doesn't include a direct response from Anthropic or SK Telecom on the legal question, so don't read this as a confirmed violation yet.

The tension in this poll is the story: 63% say AI is advancing too fast, only 16% see a positive societal impact, yet 49% of Americans are already using chatbots and ChatGPT usage doubled to 44% since 2023. Younger users are the heaviest adopters and the most pessimistic—66% of 18–29-year-olds have used one, but 48% think AI will harm society. The 30–49 crowd is the daily workhorse: 34% use a chatbot at least daily, and about 40% use AI for work tasks. The number I'd anchor on is the 66% who worry about AI spreading misinformation—that's the trust ceiling these products are hitting right now.

The reason this is worth a look: it's a concrete, open-source agent memory architecture that doesn't just wave at "long-term memory" but actually splits it into three Elasticsearch indices — episodic, semantic, and procedural — with hybrid retrieval and a reranker hitting 0.89 R@10 on 168 questions. Multi-tenant isolation is done through document-level security, and the post claims zero cross-tenant leaks. I'd discount it a bit because the post doesn't disclose latency or cost. A hybrid retrieval plus cross-encoder reranker pipeline could add meaningful overhead to agent response time, and we can't tell from the numbers given. The eval set is also small at 168 questions, and the task types aren't broken down — we don't know if it's mostly fact recall or procedural matching. That said, a few design choices are genuinely useful: supersession over deletion keeps an audit trail when facts contradict; decay weighting prevents old facts from outranking fresh ones; and periodic consolidation of raw events into durable facts keeps the index from bloating. Full code is on GitHub. If you're already running Elasticsearch as your agent backend, this is a drop-in starting point.

This is worth a click because OpenAI benchmarked against actual physicians, not just medical exam scores. In a blind review of 3,500 responses, doctors rated GPT-5.5 Instant higher than human-written answers on accuracy, communication, and completeness, with fewer failures like missing red flags or skipping follow-up context. Production monitors also show a 71% drop in health factuality issues over two months. I'd discount this a bit since it's an OpenAI blog post, not an independent study, and the post doesn't detail reviewer conflict-of-interest rules. But the approach is solid: have physicians define what good looks like in real conversations, then align the model to that standard. Free-tier users getting near-Thinking-model health responses is the concrete takeaway here.