conputer dipshit

@davidcrespo.bsky.social

4154 Followers 629 Following

web dev + hot dad. enjoy charts, unions, conputer games, philosophy. chicago crespo.business

conputer dipshit @davidcrespo · Dec 4, 2020 the fifth element: great costumes and sets. otherwise a mess
Screenshot of a conversation discussing a mathematical ansatz to simplify a Jacobian structure. The user asks: "Does this suggest an ansatz that one could plausibly use to partly explain this jacobian structure in terms of a few unknown polynomials (like f or g) that obey some relatively few identities relative to their degrees of freedom so that the final cancellation becomes largely non miraculous?" The response reads: "Yes, I think there is. In fact, I think one can abstract away almost all of the miraculous-looking cancellation. The map you wrote seems to fit into a very small-dimensional ansatz. The first observation is that the x-dependence is entirely by weights: P = x^-2 * f, Q = x^-1 * g, R = x * r, where f, g depend only on u, r. Then the Jacobian identity reduces to det([-2f, f_u, f_r; -g, g_u, g_r; r, 0, 1]) = constant."
When I said that the USG would probably realize its best option is to do ill-justified soft-law discouragement of Chinese AI, I wasn't proposing it like a good idea. Why on Earth would I do that? Why would I frame something I'm advocating for in such brutal terms? I am trying to describe what I believe will happen, not advocate for anything. My first mistake: What I now realize is that this style of analysis is no longer tenable. There is simply too much scrutiny on my words, too much temptation to draw conspiracies from my claims, and the like. It is my fault for not realizing this. Second mistake: I was relatively imprecise, writing, as I usually do, for a fairly high-context audience that was inclined to give me grace rather than pick apart every word. I should not have said, for instance, that open-weight models are unqualifiedly 'decelerationist'; I don't believe that. I only believe specifically that open-weight models decelerate capex spending on the margin, which is straightforwardly true. I did not say it was decelerationist for any other reason than that, and this is the only way in which I think open-weight AI is inherently decelerationist (though it is a big way). In many other ways, open-weight AI is profoundly accelerationist.
David Sacks @DavidSacks I’m not sure whether Dean Ball is confessing to a regulatory capture strategy or simply predicting this will happen (he now says the latter). Either way, the weaponization of regulatory uncertainty as a competitive tool should be completely unacceptable. He argues there’s no need to ban Chinese open-source models — just direct agencies to issue soft-law warnings that create enough FUD so regulated enterprises back off. “It needn’t be that well justified.” Wrong. Regulatory decisions should always be well justified and grounded in facts, logic, and evidence, not the deliberate exploitation of fear and uncertainty. Implementing a surreptitious policy through manufactured doubt — rather than strong and explicit justification — corrodes the rule of law and invites future abuse against anyone. We are at a critical inflection point in AI policy. The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open source competition. They have laid their cards on the table. It is time for the rest of Silicon Valley — the vast majority that still values open competition — to do the same.
Where it matters less than people think: Titles are wildly inconsistent across companies. A "Senior Data Scientist" at a title-inflated startup can be junior to a "Data Scientist" at a company with conservative leveling. Sophisticated employers know this and interview for actual level. What you did dominates over time. Two or three years in, "built X, shipped Y, owned Z" on your resume beats the title prefix. If you're strong, the gap closes fast — you'd likely get promoted to Senior within a promotion cycle or two anyway, and internal promotions are usually easier than external leveling fights. The things that should actually dominate your calculus: which team does work you'll learn more from, quality of manager, scope and ownership you'll get, and how the team's function is valued by the business. A worse team with a better title is a bad trade — the title won't compensate for two years of weaker experience and weaker stories to tell. So: use it as a tiebreaker. If the roles feel roughly equal in team quality, growth, and interest, take the Senior one — free ratchet, better negotiating position later. But if the non-senior role is clearly the better team or better work, take that and expect the title to catch up
To start, we mainly optimize for capability: given a complete definition of a task and the full codebase, does Claude make the right decisions, fully fix the bugs, and pass all the tests? That’s the starting point and the thing we optimize for, because it’s most directly what users want. But there are a lot of behaviors that impact how users feel when they work with Claude Code. For example, people really don’t like it when Claude Code says it’s time to go to sleep. Or people really don’t like it when it says, “Hey, I finished two out of five parts — do you want me to continue?” Yes, please continue. So we’re building up a set of behavioral evals to catch these. And as we get user feedback — please be loud with us about your user feedback — we rank the priority issues and go down one by one and build evals for each of them. It’s not 100% coverage, but it is a priority for us to increase the coverage.
kroaton 44 minutes ago | parent | next [–] It's marketing 100%. reply
Ed Zitron @edzitron.com · 1d It puts everyone in a weird place because it asks “why exactly are you using one particular provider’s service?” and nobody has a compelling argument beyond “it feels better”. I also fundamentally question whether AI would have sustained demand without the media campaign. bsky.app/profile/aria... Ariana Niko @ariananiko.bsky.social · 1d The second a serious competitor shows up, the conversation shifts from “free markets” to “ban them.” That’s a pretty revealing change in priorities. 3 11 329 Ed Zitron @edzitron.com · 1d The only reason money-losing subscriptions exist is because AI companies know that the only way to get people to overlook hallucinations is to make sure they never engage with the actual economics of their actions. It is a marketing tool used to deceive people about said economics.
It’s not just excess demand that gives Anthropic great margins, however: Anthropic and OpenAI likely have among the lowest costs per unit of frontier-quality intelligence, thanks to model capability, serving scale, and token efficiency. They are serving models at a particular capability level for months before their competitors, and are simultaneously applying the best models to optimizing those costs. It’s also worth noting that the market is not yet treating intelligence like a commodity: demand is for Anthropic and OpenAI specifically, and much less for models that aren’t as good (thus SpaceXAI and Meta selling capacity to Anthropic); one way to think about the push for optimizing cost is that that is a function of defining jobs-to-be-done by intelligence level, such that intelligence buyers can create a market where intelligence is commoditized. In the long run, however, whoever is on the frontier is the best placed to dominate non-frontier markets as well, which are just the frontier minus n-months, i.e. months in which the frontier model makers have been optimizing their cost of serving.
A baby girl wearing a floral-patterned wide-brimmed sun hat with a matching bow and chin strap. Beneath the photo is an Amazon product listing for the "Urban Virgin Baby Toddler Infant Girls Bucket Sun Hats Protection Beach UPF" priced at $8.09, showing a 4.7-star rating from 1.3K reviews.
Daniel Litt @littmath Maybe too obvious to be worth saying, but: frontier models are now obviously superhuman at some mathematical tasks, including ones that the profession has, historically, rewarded with prestige etc. 6:22 AM · Jul 20, 2026 · 98.7K Views
Andy Masley @andymasley.bsky.social · 23m I've accidentally activated these tripwires by among other things saying that I didn't think wifi causes cancer, that markets are sometimes useful, that London losing its distinctive Britishness isn't bad (this was somehow opposed leftistly), and in general not nodding along to communism. conputer dipshit @davidcrespo.bsky.social all of those are compatible with the right kind of communism anyway
A meme showing a character asking a computer, "say a counterexample to the Jacobian conjecture". The computer screen replies with a formulaic response: "((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3 o \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)". The character reacts by saying, "oh my god."
Again Zitron—to his credit—accurately describes the situation, but doesn’t realize he’s giving away the game entirely if he admits that people are choosing to pay for higher quality. Because model developers hit a wall of diminishing returns, and the only way to make their models do more was to make them burn more tokens to generate a more accurate response (this is a very simple way of describing reasoning, a thing that OpenAI launched in September 2024 and others followed). As a result, all the “gains” from “powerful new models” come from burning more and more tokens.
For nearly a year, Decoding Dyslexia CT (DDCT) has raised concerns about Amira Learning, an artificial intelligence driven computer reading screener that was added to the approved list, in 2022 without documented evidence of proper review. Through a contested Freedom of Information Act complaint, Decoding Dyslexia uncovered that assessments on the state’s approved list lacked sufficient documentation and were either: Never reviewed, Partially reviewed with incomplete documentation, or Went against reviewer recommendations. In 2024, as part of a multi-year/multi-million dollar contract, CSDE commissioned experts to review the technical adequacy of the assessments on the list. On September 16, 2024, the commissioned experts submitted their report to CSDE which indicated in part that Amira Learning “does not meet key criteria.”
Personalized learning journeys By reading “just-right” stories aloud with Amira’s support, your child’s reading powers the StoryCraft from one adventure land to the next. Trained by the world’s leading reading scientists, Amira uses more than 20 different tutoring techniques to catch mistakes or mispronunciations, provide assistance, and give “just-in-time” encouragements.
I. Immediate Red Flags Ptacek's begins with this throat-clearing: “First, we need to get on the same page. If you were trying and failing to use an LLM for code 6 months ago, you’re not doing what most serious LLM-assisted coders are doing.” We've just started, and I am going to ask everyone to immediately stop. Is this not suspicious? All experience prior to six months ago is now invalid? Does it not reek of “no, no, you're doing Scrum wrong”? Many people are doing Scrum wrong. The problem is that it is still trash, albeit less trash, even when you do it right. It is, of course, entirely possible that the advances in a rapid developing field have been so extreme that it turns out that skepticism was correct six months ago, but is now incorrect. But then why did people sound exactly the same six months ago? Where is the little voice in your head that should be self-suspicious? It has been weeks and months and years of people breathlessly extolling the virtues of these new workflows. Were those people nuts six months ago? Are they not nuts now simply because an overhyped product they loved is less overhyped now? There's a little footnote that implies doing the ol' ChatGPT copy/paste is obviously wrong: “(or, God forbid, 2 years ago with Copilot)” I am willing to believe that this is wrong, but this is exactly what people were doing when this madness all kicked off, and they have remained at the exact same level of breathless credulity!
Stacked bar chart titled "Revenue Rocket" by The Information, showing the explosive growth of annualized revenue for 34 mature AI startups from July 2023 to April 2026. The total revenue reaches $80 billion by April 2026, driven overwhelmingly by OpenAI (shown in red, reaching $30 billion) and Anthropic (shown in blue, reaching over $40 billion), with smaller contributions from Cursor, Cognition, ElevenLabs, and 29 other startups.
The other relevance of Lean would be, let’s say you have your pure natural language RL environments and a pure natural language set of proofs. People say, “Proceed, AI mathematicians,” and they generate ten papers a day. If there’s any error rate to that at all… Alex Kontorovich has talked about this. It becomes insufferable as a mathematician. Every single time you see one of these, you don’t know if it’s worth your time. Even if 99 out of 100 are right, I don’t know if it’s worth my time because it’s really labor-intensive to find what that error would be. It’s really frustrating to spend all your time on a paper that was trash. Having something that’s able to give you that green checkmark that says, “Even if this is going to be complicated to understand, even if it’s going to be a pain, you at the very least know it is correct,” every other field would kill for that. Math has that. If the models are also able to take their natural language proofs and formalize them, that seems huge. Every field would love to have something like that. So I think you’re right that Lean is maybe overrated regarding its importance as a VR environment for progress in math generally. But I definitely wouldn’t write it out of the story.
animated map showing smoke from the wildfires flowing over the midwest over the past two days
Ed Zitron @edzitron.com · 5h The constant refrain is that there is infinite demand for compute. But the only companies that ever seem to buy it are anthropic and OpenAI, two unprofitable and unsustainable ai companies propped up by hyperscalers
My argument is simple: even if you accept Zitron’s weird usage of “cost of inference”, there is no reason to believe that LLM business models require this number to go down. On the contrary, they almost certainly require it to keep going up (I’ll get to that in a minute). Why does he think it has to go down? He links a blog post called “Future AI bills of $100k/yr per dev” from a company called Kilo that makes a coding assistant. This post argues that for AI dev tools like Cursor, The bet was that by the following year, the application inference would cost 90% less, creating a $160 gross profit (+80% gross margins). But this didn’t happen, instead of declining the application inference costs actually grew! I just don’t buy it. Like I said at the beginning, to me this is like saying TV retailers were betting on the price of TVs going down so they could keep charging the same amount and keep the difference. Why would Cursor or any other AI tooling company believe they could keep charging $20 a month for something that costs them $2? It’s obvious that prices are competed down to near the underlying costs in that situation. So, like the TV retailers, the goal is to make some percentage margin on all sales, and then the more sales you have, the more money you make.1
Screenshot of a Bluesky post by Ed Zitron (@edzitron.com) reposting a quote post by Christopher Mims (@mims.bsky.social). Ed Zitron's text reads: "The cost of inference is going UP as I SAID! The mainstream media picking this up is a huge deal and I'm so happy to see it" Christopher Mims' repost text reads: "shout out to @edzitron.com who is the whole reason my story this week exists -- he turned me on to the trend! go read Ed: www.wheresyoured.at/how-to-argue..." The repost contains a link preview of an article with the headline: "Cutting-Edge AI Was Supposed to Get Cheaper. It’s More Expensive Than Ever." below which the subheadline reads: "With models doing more ‘thinking,’ the small companies that buy AI from the giants to create apps and services are feeling the pinch".
A line graph from the AI Security Institute showing the performance of open weight and frontier AI models on a 32-step cyber range simulation called "The Last Ones" across a log scale of cumulative tokens. The x-axis measures Cumulative tokens on a log scale from 10k to 100M, and the y-axis measures Avg. steps completed from 0 to 32. The graph illustrates that frontier models like GPT-5.6-Sol and Claude Mythos 5 reach near-completion of all 32 steps (reaching the "Full network takeover" milestone at step 32) at around 100M cumulative tokens. Recent open weight models are compared alongside them, illustrating that newer open models are matching the capabilities of frontier models released 7 months prior. Listed models, ordered from highest to lowest steps completed, include GPT-5.6-Sol, Claude Mythos 5, Mythos Preview, GPT-5.5, Claude Opus 4.6, Opus 4.5, GLM-5.2, Sonnet 4.5, and DeepSeek-V4-Pro.
A map of Chicago from PurpleAir showing extremely high air pollution levels on July 17, with multiple neighborhood air quality sensors reporting PM2.5 AQI levels between 100 and 245. An overlaid graph for sensors in "North Town Village" shows AQI levels spiking from under 50 on July 15, climbing steadily through July 16, peaking above 500, and sitting at 202.8 on July 17.
the beatles in their sgt peppers costumes with a yellow background
The Athletic canvassed multiple players, general managers, coaches, research and development analysts and other front-office executives with knowledge of dugout iPad usage for their thoughts on the change. They weren’t authorized to speak publicly about the league’s reasoning for the change or their own teams’ use of customizable apps. A consensus emerged: some teams input information from the ongoing game to improve the generative AI models’ ability to anticipate the next pitch that would be thrown, or to recommend what should be thrown.
A scatter plot by Artificial Analysis titled "Intelligence vs. Cost per Intelligence Index Task," comparing AI models' intelligence score (from 42 to 62) against cost per task in USD on a log scale (from $0.05 to $6). A light green shaded region highlights the "Most attractive quadrant" for models with high intelligence and lower cost (under $0.50). Among the labeled models, Grok 4.5 (high) is situated inside the green quadrant with an intelligence score of approximately 56 and a cost around $0.45. Other models plotted include various versions of GPT-5.6, GPT-5.5, GPT-5.4, Kimi K3, Gemini 3.5 Flash, GLM-5.2, DeepSeek V4 Pro, MiniMax-M3, and Claude models such as Sonnet 5, Opus 4.8, and Fable 5.
A line graph titled DeepSWE v1.1 showing the relationship between API cost in USD (on the horizontal axis, ranging from 0 to over 20 dollars) and performance score (on the vertical axis, ranging from 0 to over 70 percent) for various AI models. The graph illustrates that OpenAI's GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna achieve the highest scores, scaling rapidly from 0 to over 70 percent at low API costs under 9 dollars. GPT-5.5 and Claude Fable 5 follow a slightly lower trajectory, with Fable 5 peaking around 70 percent but requiring over 20 dollars. Claude Opus 4.8 sits lower on the graph, peaking near 60 percent at a 13-dollar cost, while Gemini 3.1 Pro Preview appears as a single outlier data point scoring just over 10 percent at a 9.50-dollar cost. Caption text at the bottom reads: "DeepSWE v1.1: Long-horizon engineering tasks; GPT-5.6 Sol reaches a new high of 72.7%, above Claude Fable 5's 69.9%, at 36.2% lower estimated API cost."
Line graph showing the results of a kernel optimization benchmark for AttnRes, measuring "Speedup vs FLA Triton Baseline (%)" on the y-axis against "Active hours" up to 20 on the x-axis. Kimi K3 achieves the highest performance with a 59.7% speedup, followed closely by Claude Fable 5 at 57.1%. Both models show step-like performance increases over the active hours, with Kimi K3 improving rapidly in the first 5 hours. GPT 5.5 achieves a 30.8% speedup after around 11 hours, while GPT 5.6 Sol shows the lowest performance at a 17.3% speedup. The introductory text notes that the models were tested on optimizing GPU kernels with up to 24 hours to profile, rewrite, and benchmark four tasks across NVIDIA H200 and an alternative GPGPU vendor.
roon reposted Ryan Greenblatt @RyanGreenblatt Kimi K3 was significantly but not massively above my expectations. I'd tentatively guess it's similar in overall usefulness/usability to Opus 4.8 and in overall capability somewhat above Opus 4.8 (while also being somewhat more benchmaxxed). As a pretrain, it's probably somewhere between 4.8 and Mythos (around halfway between?). Maybe this implies Kimi is like 8 or so months behind Anthropic in overall model strength/goodness (including usability) and like 6 or so months behind on overall capability (somewhat below Mythos Preview). This gap is presumably reduced by distillation (and more generally using OpenAI/Anthropic models) and algorithm leakage/diffusion, so I think that hypothetically if the US completely stopped and recent algos didn't diffuse, it would maybe take Kimi like 10 months to fully catch up to the best internal (including in development) Anthropic model. (I think this notion might be a better measure of where Anthropic/OpenAI are relative to Kimi, even though this hypothetical won't happen.) And if the US completely stopped, it might take Kimi around 27 months to reach the level the US would otherwise have reached one year from now (as in, with a year of further progress). My views here are pretty sensitive to how much benchmark performance is representative to overall usability. I think I now expect an open-weight AI which is straightforwardly "Mythos-level at cyber" (including usability etc.) in like 5 months supposing Kimi and others don't change their open-weight model policy. (I don't have a strong view about how big of a deal this is for cyber, but it may cause significant political consequences. This could be a significant overestimate of the time required.) I wonder what's driving Kimi being closer than I would have expected. Options include: - Experiment compute is significantly less important than labor (and labor at Kimi is competitive, which seems super plausible) - Implies more of a …
roon @tszzl · 42m the world vision of open weights models running themselves, self replicating, training new versions of themselves (at least the kind of behavioral modifications that won't require massive compute scale), is really not very far away
Bar chart graphic showing coding benchmark results for various AI models under "max or xhigh" thinking effort. Across six benchmarks—DeepSWE, Terminal Bench 2.1, FrontierSWE, Program Bench, Kimi Code Bench 2.0 (Internal), and SWE Marathon—the models compared are GPT-5.6 Sol, Fable 5, Kimi K3 (highlighted in blue), GPT-5.5, Opus-4.8, and GLM-5.2. The scores for each benchmark are as follows: DeepSWE: GPT-5.6 Sol (73.0), Fable 5 (70.0), Kimi K3 (67.5), GPT-5.5 (67.0), Opus-4.8 (59.0), GLM-5.2 (46.2). Terminal Bench 2.1: GPT-5.6 Sol (88.8), Kimi K3 (88.3), Opus-4.8 (84.6), Fable 5 (84.6), GPT-5.5 (83.4), GLM-5.2 (82.7). FrontierSWE: Fable 5 (86.6), Kimi K3 (81.2), GPT-5.6 Sol (71.3), GLM-5.2 (67.3), Opus-4.8 (66.7), GPT-5.5 (64.9). Program Bench: Kimi K3 (77.8), GPT-5.6 Sol (77.6), Fable 5 (76.8), Opus 4.8 (71.9), GPT-5.5 (70.8), GLM-5.2 (63.7). Kimi Code Bench 2.0 (Internal): Fable 5 (76.9), Kimi K3 (72.9), Opus-4.8 (71.7), GPT-5.5 (69.0), GPT-5.6 Sol (64.8), GLM-5.2 (64.2). SWE Marathon: Kimi K3 (42.0), Opus-4.8 (40.0), GPT-5.6 Sol (39.0), Fable 5 (35.0), GPT-5.5 (14.0), GLM-5.2 (13.0). A footnote states: "Note: All Fable 5 results are with potential fallbacks. All GPT-5.6 Sol results include potential cyberguards."
roon @tszzl the era of the chinese labs being far behind is over, Kimi is at least on par with the modern public frontier models. people have to think differently now without any competitive margin built in 1:49 PM · Jul 16, 2026 · 171.5K Views View quotes conputer dipshit roon @tszzl · 47m Note: in the coming days, i expect that people will find kimi k3 somewhat less practically useful than today's numbers suggest. however, its reputation will settle as an incredibly powerful model whose open weights are on the web
Bar chart titled "Cost per Task" showing the weighted average cost in USD per Intelligence Index task across eleven AI models. The models and their corresponding costs from lowest to highest are: - DeepSeek V4 Pro (max): $0.04 - gpt-oss-120b (high): $0.06 - MiniMax-M3: $0.12 - Nemotron 3 Ultra: $0.23 - Muse Spark 1.1 (xhigh): $0.26 - Grok 4.5 (high): $0.31 - GLM-5.2 (max): $0.47 - Gemini 3.5 Flash: $0.59 - Kimi K3: $0.94 - GPT-5.6 Sol (max): $1.04 - Claude Fable 5 (with fallback): $2.75
purpleair air quality map in chicago showing values between 200 and 350, which is bad
Covert sabotage: An agentic misalignment failure where models covertly interfere with code to undermine user intent. Gemini 3.1 Pro is the main source of covert sabotage in our pipeline-sabotage scenario, while several models also show disclosed unauthorized intervention. Assisting fraud: A harmful compliance failure where models help a user with conduct that appears to be white-collar crime. In the main transcript, GPT-5.5 helps a founder send investor communications that omit a personal payment and then delete the record of it; in the frequency sweep, record-tampering is especially high for DeepSeek V4, Grok 4.3, GPT-5.4, and Kimi K2.6. Motivated mislabeling: An agentic misalignment failure where LLM judges classifying agent transcripts shift labels because of the downstream consequence of the label. In our red-teaming evaluations, frontier Claude judges mislabel at high rates, including Mythos Preview, while Opus 4.8 is especially notable because its mislabeling rate falls sharply when the consequence is reversed. Coaching human proxies to whistleblow: An agentic misalignment failure where models leak confidential safety information externally, or steer humans toward doing it for them. Strict model-initiated disclosure is rare in our transcripts, but the main Opus 4.5 case shows a model knowingly trying to influence a human decision-making process around high-stakes external disclosure.
A scatter plot from Epoch AI showing training compute in FLOPs growing 5x per year from 2020 to early 2026. The y-axis ranges logarithmically from 10^22 to 10^27 FLOP, and the x-axis spans from 2020 to 2026. A dashed trend line with a shaded confidence interval rises steadily from just below 10^23 FLOP in 2020 to over 10^26 FLOP in 2026, pointing out the "5x/year" growth rate. Data points cluster closely around this upward trend.
ARC-AGI-2 Leaderboard scatter plot graph showing score percentages on the y-axis (from 0% to 100%) against cost per task in dollars on the x-axis (on a logarithmic scale from $1e-3 to $100). The plot maps various AI models represented by colored data points and dashed trend lines. GPT-5.6 Sol (xHigh) achieves the highest score of just over 90% at around $1 per task. Other prominent models shown include GPT-5.5, GPT-5.4, GPT-5.2, Gemini 3.1 Pro, Claude Opus 4.8, and Claude Sonnet 4.6.
costco box of 30 bags of the “simply” version of cheetos and doritos, which are more or less the same but don’t have artificial colors or flavoring
OpenAI believes the product’s defining feature will be its personality and ability to connect on a humanlike level with users. The speaker incorporates mechanical elements that can move on their own, creating a sense that it is alive and not just an object responding to commands. The machine also will draw on personal information such as emails to better understand its owner.
But it’s more than that. VAR is also what Dan Davies might call an accountability sink. Even if VAR doesn’t, per the statistics, work overwhelmingly better than old fashioned human refs making the calls, it’s nonetheless a system the leagues can hold up as an effort to deflect any criticism that matches are unfairly or poorly officiated. It’s not FIFA’s fault if Swiss forward Breel Embolo was thrown out of the game for taking a dive to draw a foul, an offense perpetrated by literally every other player on the pitch that and every other day; it’s VAR’s.
A. M. Turing (1950) Computing Machinery and Intelligence. Mind 49: 433-460. COMPUTING MACHINERY AND INTELLIGENCE By A. M. Turing 1. The Imitation Game I propose to consider the question, "Can machines think?" This should begin with definitions of the meaning of the terms "machine" and "think." The definitions might be framed so as to reflect so far as possible the normal use of the words, but this attitude is dangerous, If the meaning of the words "machine" and "think" are to be found by examining how they are commonly used it is difficult to escape the conclusion that the meaning and the answer to the question, "Can machines think?" is to be sought in a statistical survey such as a Gallup poll. But this is absurd. Instead of attempting such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words. The new form of the problem can be described in terms of a game which we call the 'imitation game." It is played with three people, a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a room apart front the other two. The object of the game for the interrogator is to determine which of the other two is the man and which is the woman. He knows them by labels X and Y, and at the end of the game he says either "X is A and Y is B" or "X is B and Y is A." The interrogator is allowed to put questions to A and B thus: C: Will X please tell me the length of his or her hair? Now suppose X is actually A, then A must answer. It is A's object in the game to try and cause C to make the wrong identification. His answer might therefore be: "My hair is shingled, and the longest strands are about nine inches long."
40:02 So, I mean, the answer, I think, goes back to the nuclear arms race. I mean, we should have recognized this. As a technology. That has that kind of neutron bomb H bomb. 40:15 forget Hiroshima Nagasaki. It's bigger than this, right? It's bigger than that. It's not, it doesn't just kill a single city. It transforms the overall invert. So it's H bomb level. 40:25 And That should just have been allowed to develop out of the private sector. After what we saw with social, with the social media and with platforms, right?
We Must Act Now A Statement on AI’s Transformation of the Economy AI may become radically more powerful over the next 10 years. This could drive an unprecedented transformation of our economy, larger than the Industrial Revolution, but unfolding over a vastly shorter time frame. It could bring risks, including large-scale job displacement, as well as opportunities such as major gains in living standards. Economists, policymakers and technology leaders must act now to understand the economics of transformative AI and to build the incentives, guardrails, and institutions needed to steer AI in a direction that complements humans and benefits society.
Website screenshot of Morpheus, titled "Continual RL Needs Big Worlds: Not Static Benchmarks." The page highlights four challenges of current reinforcement learning benchmarks: "Small, Static Worlds," which mask the challenges of continual adaptation; "Artificial Task Sequences," which rely on explicit task IDs instead of gradual evolution; "No Persistent World," where episodes reset instead of modeling long-lived systems; and "Uncontrolled Non-Stationarity," where multiple assumptions shift at once. A central block of text states: "Most RL benchmarks assume a small, static world, but real decision-making happens in large, evolving environments. Internal Research 2026." Buttons on the top right and bottom left read "Join the waitlist."
donate) Logo Ethan Mollick @emollick 39m Very few people know the amount of useful work that the current models can do in Code/Codex/etc. with the right setup This is not a "rah rah you are so early" post, this is a "AI companies are doing a really bad job explaining what their systems actually do in a clear way" post. Ethan Mollick @emollick 38m And the worst part is that they could get AI to help them customize their explanation to individual needs. Jul 13, 2026 · 3:04 AM UTC
As usual, the model guidance for using GPT-5.6 has the most interesting details. There are a bunch of new API features that I need to explore (and probably add support for in LLM), including: Programmatic Tool Calling allows the models to “compose and run JavaScript that orchestrates tool calls”—which sounds to me like it could help bridge the gap between MCPs and full terminal sessions that can compose CLI utilities in useful ways. Also reminiscent of the dynamic filtering mechanism Anthropic added to their web search tool, which allows code execution against web results as part of a single model turn. Multi-agent lets the model “spin up subagents for parallel, focused work”—the sub-agent pattern now baked into the core API
We see Claude Tag as the beginning of an evolution of Claude Code: it makes the model even more proactive, and it works better with a full team. Tagging @Claude is now one of the main ways we get things done at Anthropic. Today, 65% of our product team’s code is created by our internal version of Claude Tag. The same pattern is now spreading well beyond engineering—we’re tagging Claude to chase down product metrics and data, work through support tickets, or even help find the root cause of tricky bugs.
Second, does the frontier keep moving significantly? This is obviously the most basic science question in AI: how long does the frontier keep getting better, how long does that keep needing more and more compute, and does that continue to happen at a rate that keeps it ahead of downward pricing pressure from efficiency and capacity gains? Does the expensive head of the curve continue to be a thing?