posts · 2026-06-16

Wolfram Language and Mathematica 15 dropped, and both HN and AI media are pointing to Stephen Wolfram's own blog post. The coverage is consistent because there's only one source—no third-party reviews or independent testing yet. The headline feature is a built-in AI assistant inside every notebook. The thing that makes it different from a generic chatbot: it can actually invoke Wolfram's symbolic computation engine. Ask it to solve an equation, and it doesn't just generate code—it runs the computation. The reverse direction works too: you can call Wolfram tools from external AI environments like LLMs. Other updates: symbolic music (representing and generating musical structures in Wolfram Language), expanded time series and categorical data support, a unified ModelFit function, and notebook UI improvements like sidebars and dark mode. I'd take this with a grain of salt for now. We only have Wolfram's own announcement—no pricing details, no user benchmarks, no performance comparisons against version 14. The blog doesn't specify which underlying model powers the AI assistant or whether it costs extra.

This one's worth opening because Ferriss is the poster child for how-to nonfiction—The 4-Hour Workweek has been selling for over a decade. He now says his own print sales are "dizzying" in decline and cites Publishers Weekly: adult nonfiction dropped 9% in Q1 2026, with self-help plunging 26.3%, the steepest subcategory fall. His argument is straightforward: readers now ask ChatGPT or Claude instead of buying a book for step-by-step advice. I'd discount the personal evidence, though. He never shares a single unit number or year-over-year percentage for any of his five books. He says "I've included the spreadsheet below," but the actual post doesn't contain it. Without figures, it's just an anecdote from a worried author. The industry data is real, but self-help declines could also reflect publishing cycles or post-pandemic reading shifts—Ferriss himself admits "other factors may be at play." If you're in publishing or paid knowledge products, the useful bit is the direction, not the conclusion. LLMs are definitely eating the "quick how-to lookup" use case. But books as deep frameworks aren't dead yet—just the surface-level instruction manual is getting replaced.

The 41% vs 39.5% headline is eye-catching, but I'd discount it a bit: the data comes from Ramp, a spend management platform, so it only reflects subscription distribution among its own customers—not the entire enterprise market. Ramp's customer base skews tech and startups, where Anthropic already has strong traction. The more interesting claim is that controversy drove adoption. Ramp's chief economist says a similar incident in March—when the DoD flagged Anthropic as a supply-chain risk—led to record enterprise uptake, with spending concentrated on Claude Opus 4.8. It's the TikTok-ban-download-spike logic: the fight itself becomes free exposure. That said, the post doesn't give absolute dollar amounts or customer counts, just share shifts. Anthropic just closed a $65B round at a $965B valuation and confidentially filed for IPO. Data dropping at this moment—I'd keep some skepticism handy.

This one's worth a look because it changes how you pay for experimentation on Midjourney. Before, every generation burned standard fast hours. Draft mode spits out 24 low-res previews at half the cost—pick the ones you like, hit Vary, and only then pay full price for the high-res render. If you iterate a lot, that roughly halves your exploration spend. The --preview flag is a separate thing: an early-access channel for unfinished models. Midjourney is upfront that outputs may be rough and jobs aren't guaranteed to stay consistent, especially with personalization and moodboards. Treat it as a public beta, not a stable feature. The post doesn't disclose Draft mode's exact resolution or which model --preview points to. I'd hold off on those two details until they're clarified.

This piece is worth opening because Marcus lines up three data points that hit at the same time. First, OpenAI's market share dropped below 50% for the first time, with Google eating into it. Regular users don't see a difference between ChatGPT and Gemini, and pure LLM businesses lack stickiness. Second, Microsoft is exploring DeepSeek over OpenAI for Copilot because usage-based pricing makes costs too high—when your biggest backer starts shopping for alternatives, that's a hard signal. Third, Ed Zitron got audited 2025 financials: $13.07B revenue against $34B in costs, losses up nearly 8x year-over-year. Marcus also notes Washington might accidentally help OpenAI by targeting Anthropic with export controls, but he stands by his prediction that OpenAI gets acquired, with Elon Musk as a dark-horse bidder. I'd discount this a bit: Marcus has been consistently bearish on OpenAI, and he's picking data points that fit that view. The market share chart doesn't show methodology or source, the Microsoft-DeepSeek story is a single tweet, and the financials come from Zitron's reporting on audited documents. Reads more like a well-timed opinion roundup than new reporting.

The reason to click: the sender is the US Commerce Secretary and the recipient is Anthropic, maker of Claude. If the US government does restrict "the most advanced AI models," Claude Sonnet 4.5 and any future Claude 5 would almost certainly be on the list. But right now we only have the title and recipient—the letter's actual demands, whether it targets model weights or API access, and the timeline are all missing from the post. I'd treat this as a policy signal, not a reason to start modeling Anthropic's revenue impact.

The letter itself is more revealing than the headline. Lutnick didn't use vague national-security language — he directly cited Mythos generating deepfakes of Trump and Musk, and demanded Anthropic explain why its content moderation failed to catch them. Anthropic's move was voluntary, not a government-ordered takedown. Two things the article doesn't answer: whether this is temporary or permanent, and whether Anthropic gave any timeline for restoration. The body just says "disabled." I'd read this as a regulatory pressure case, not a product death sentence — Mythos's text capabilities are still live, only image gen got pulled.

Gergely Orosz lays out a clear timeline of Meta's engineering org unraveling since April. For two decades, the culture evolved from 'move fast and break things' to 'move fast with stable infra,' with engineers empowered to balance impact and quality. Then leadership flipped to an AI-first mandate that treats core engineers as disposable. The post details forced data-labeling assignments, what Orosz calls Meta's most embarrassing outage, and internal process breakdowns. I'd read this as a signal: when a company redefines engineering from profit center to cost center, the culture can collapse fast. The article doesn't include an official Meta response to these changes.

The real signal here isn't the model choice — it's the pricing collapse. Copilot Cowork's unlimited plan got crushed by users running hundreds of tasks per week, forcing Microsoft to switch to usage-based billing. The DeepSeek V4 option is a cost play: a fine-tuned, safety-guarded version hosted on Azure, already working but not yet committed. I'd discount the "Microsoft adopts DeepSeek" framing — Axios didn't share pricing, task types, or latency numbers. For now, read it as cost pressure, not a strategy shift.

This caught my eye because Ferriss rarely shares his own sales data. His five books show a steady domestic print decline from 2023 to 2026, and Publishers Weekly data backs it up: adult nonfiction fell 9% in Q1 2026, with self-help plunging 26.3% year-over-year. Ferriss's take is blunt—readers now get answers from tools like Claude instead of buying a 300-page how-to book. I'd discount this a bit. He didn't disclose exact unit numbers per title, so we can only see the direction, not the magnitude. And his books—The 4-Hour Workweek, Tools of Titans—are essentially method compilations, exactly the type of content AI answers best. Biography, investigative journalism, and narrative nonfiction probably aren't getting hit this hard. Still, a 26.3% drop is real. If this holds, the self-help business model—package a methodology into a $15-25 book—gets squeezed hard by free, instant AI answers. Publishers will likely push authors toward courses, communities, and paid newsletters, with the book becoming a lead magnet rather than the main product.

This piece is worth reading because it dismantles the official narrative from last Friday's ban. The Commerce Department invoked export controls to block non-US persons from accessing Fable 5 and Mythos 5—including Anthropic's own foreign staff. Anthropic suggested the trigger was a jailbreak that crossed a safety line, but TechCrunch points out the government has produced zero technical details or jailbreak evidence. The article's read is that this looks more like a reactionary political move than a technical regulatory action. I think that holds up: if there were a real jailbreak report, we'd typically see a technical summary or red-team findings attached, not just a vague 'national security' label. That's the missing piece right now.

This piece is worth opening because Gergely connects the dots on two chaotic months inside Meta. The core claim: leadership now treats software engineering as a cost center, not a profit driver. The most concrete detail is engineers being forcibly assigned to AI data-labeling work, which tanked morale and preceded a major outage. He contrasts this with the 2012 'little red book' culture and his 2022 deep-dive, but the article relies on external observation and insider accounts—no internal decision-making details are disclosed. I'd discount it slightly: Gergely's lens skews toward engineer experience and gives less weight to business pressures. Still, if the forced-labeling claim holds, it signals that AI anxiety inside Meta has overridden respect for engineering quality.

The reason to click: Zhipu shipped three things at once — genuinely usable 1M-token lossless context, MIT license for commercial use, and Day 0 inference support on domestic hardware including Huawei Ascend. It scored 51 on the Artificial Analysis composite leaderboard, top three with Anthropic and OpenAI, and hit #1 among globally available models in the Code Arena front-end dev blind test. I'd discount the benchmarks a bit. The blog picks favorable comparisons: only 1% behind Claude Opus 4.8 on FrontierSWE, but 13% behind on SWE-Marathon — that gap is real. Still, the 880K-token multi-platform app demo, if unedited, suggests the 1M context isn't just a spec-sheet number. Developer feedback points to two things: more stable project-level context, and fewer derailments on long tasks. That matters more for actual work than leaderboard scores. IndexShare keeps FLOPs at 2.9× at 1M context, and the improved speculative decoding helps too — inference cost should be lower than naive context extension. What's missing: API pricing and throughput numbers. Those decide whether this goes to production.

The reason this matters: it changes the cost structure of long-context models. Standard attention scales quadratically — double the text, quadruple the compute. SubQ's SSA routes attention based on content relevance instead, hitting 64.5x less compute than dense attention at 1M tokens. The long-context retrieval numbers are genuinely strong: near-perfect needle-in-a-haystack from 1M to 12M tokens, 99.12% on RULER at 128K. General reasoning holds up too — 85.4% on GPQA Diamond and 89.7% pass@4 on LiveCodeBench put it in the mid-tier frontier range. A few things I'd discount. The post doesn't name the base open-weight model they started from, so we can't separate what came from the original model vs. SSA and continued pretraining. The 12M-token NIAH result is impressive, but training was mostly at 1M — the 12x generalization relies on content-based routing rather than positional patterns, and I'd want to see more task diversity before calling it solved. AutomationBench Finance at 13% is still well below GPT-5.5's 18%, so the long-context advantage hasn't fully translated to complex agent tasks yet. Right now it's only deploying with design partners. The broader 2M–12M token lineup is planned for later this year. If inference costs really stay linear, this could accelerate long-document agents and full-codebase reasoning significantly.

This post lands because Vicki Boykis isn't benchmarking—she's been running local models for real work over a year. Her take: Gemma 4 26B and the newer 12B qat variant, paired with Pi for agentic coding, hit roughly 75% of frontier-model accuracy. The tasks aren't flashy—refactoring scripts, writing unit tests, scaffolding a recommender—but she says local models couldn't do any of this six months ago. I'd discount two things. She doesn't share token speeds, only that the KV cache can eat all 64GB of RAM, so actual latency might still be rough. And the 75% figure is her personal vibe metric, not a standardized eval. That said, she runs everything inside a permission-restricted Docker container, which already filters out a lot of safety headaches—useful blueprint if you want to try local agents yourself. If you've got a 64GB Mac, this is worth replicating. If not, I'd wait for more real-world reports on the 12B qat before buying hardware.

This is worth a look because Xiaomi turned an agent product into something non-developers can actually use—no local setup, just a browser. MiMo Claw ships with MiMo-V2.5-Pro, natively adapted to the OpenClaw framework and MCP protocol. The practical hook is deep WPS Office integration: it generates and edits Word, Excel, PPT, and PDF files online, which matters for domestic office workflows. Xiaomi claims a 63.8% task pass rate on ClawEval, 40–60% lower token consumption than comparable products, and roughly 3× inference throughput gains. Those are first-party numbers with no independent benchmarks, so I'd discount them a bit. The free tier gives 4 hours daily—fine for light use—and paid plans start at ¥14.9/month, which isn't expensive. What's missing: overseas subscriptions aren't open yet, and there are no third-party evals on complex multi-step tasks. Treat this as a WPS-ecosystem agent you can try with low friction, not something to measure against OpenAI or Anthropic's agent offerings yet.

This caught my eye because it bundles three things: a flagship model, a Claw product, and an office suite. The MiMo-V2.5-Pro uses MTP three-layer decoding to roughly triple OpenClaw workflow throughput. ClawEval task pass rate is 63.8%, with token consumption 40–60% lower than peers. Free daily sessions went from 1 to 4 hours, and paid plans start at ¥14.9/month. I'd discount this a bit. The source is an RSS snippet — it doesn't spell out ClawEval's test set or which peers it compared against, so 63.8% lacks context. A million-token context window and a thousand consecutive tool calls sound impressive, but we don't know if they hold up under real workloads. The practical bit is the Kingsoft Office integration: online creation and editing for Word, Excel, PPT, and PDF. That's a more concrete use case than a pure chat product, but office docs demand high format fidelity. Without real-world tests, I wouldn't get too excited yet.

The reason to click: the DOJ's defense is unusually blunt. A DOD official stated Grok is one of four models running mission-critical ops on classified networks, including recent strikes on Iran. The NAACP sued over unpermitted gas turbines at xAI's Colossus 2 site in Mississippi—turbine count jumped from 27 to 57 since April, with a 111% spike in nitrogen oxide emissions. The DOJ didn't cite a specific national security statute, but tying one company's emissions dispute directly to military operations is the real signal here. I'd discount this a bit: the post doesn't say how the judge responded or what the odds of dismissal actually look like.

This one's worth opening because Ed Zitron got audited financials that lay OpenAI's numbers bare. In 2025, $13.07B in revenue against $34B in costs — a $20.92B operating loss. R&D alone was $19.18B, with $10.59B going to Microsoft for training and cloud; total Microsoft payments hit $17.2B for the year. A $41.55B fair-value charge from the nonprofit-to-for-profit conversion pushed the net loss to $38.53B, nearly 8x the $5.09B loss in 2024. I'd discount the headline a bit: the post doesn't explain how $17.87B in costs got removed via noncontrolling interests, so don't read the $38.5B as pure cash burn. But even the operating loss is staggering — $13B in, $34B out, spending $2.60 for every dollar earned. SoftBank paid OpenAI $867M, Microsoft paid $303M; those are revenue, not investment. Don't read this as "OpenAI is doomed." The more useful take: revenue is growing, but costs are growing faster. The conversion charge masks an operating loss that was already enormous. If revenue growth can't catch up to the R&D and cloud bills, this structure gets hard to sustain on fundraising alone.

The 3-4x multiple is what makes this worth clicking. The user base is spreading beyond developers into HR, ops, and admin roles—meaning agent-based office work isn't just demos, people are actually using it daily. I'd discount it a bit though. No absolute DAU numbers, no methodology—are we counting installs, active sessions, or paid users? Meanwhile Trae Work, QoderWork, and Kimi Work are all rebranding or shipping new versions, so rankings are probably still fluid. Tencent Cloud calling this a "once-in-a-decade opportunity" is marketing noise. The real signal to watch is whether the enterprise edition and project features can move agents from solo assistants into team workflows. Paid retention data next quarter would tell us more than a DAU multiple.

The numbers are what make this worth clicking: a $60B all-stock deal for Cursor, which was about to close a $2B round at a $50B valuation. SpaceX is using its freshly-minted IPO shares to bolt an AI coding startup directly into its xAI-powered AI unit. During the roadshow, SpaceX pitched investors a $26 trillion AI addressable market — Cursor plugs the coding piece of that story. But the post doesn't spell out product integration, team retention, or regulatory approvals, and those are the parts that determine whether this actually closes. I'd discount the headline a bit: a $60B stock deal right after an IPO has a lot of moving parts around pricing and lockups.

The headline sounds dramatic, but the body only gives three MAU numbers: ChatGPT 1.1B, Gemini 662M, Claude 245M. How "below 50%" was calculated, what the measurement window is, and whether API traffic is included—none of that is disclosed. I'd discount the precision for now. The direction is real though. Gemini hit 662M through Android pre-installs and Google's ecosystem bundling; Claude holds 245M on coding reputation. ChatGPT no longer owns the room. The thing to watch is whether OpenAI's next disclosure confirms the trend, and how they respond—price cuts, enterprise bundling, or pushing agent workflows.