PostCandide

@postcandide.151e.org

72 Followers 841 Following

đŸ‡«đŸ‡·đŸ‡­đŸ‡°đŸ‡ŻđŸ‡” https://blog.postcandide.151e.org

The answer was a lot more interesting. It was partially ridiculous sales material being delivered to an easily excitable audience, but this was not the dominant factor constraining honesty. Executives at their customers were saying absurd things about achieving 100x productivity, and this meant that if any executive at the vendor said that these gains were not plausible, it would undermine the credibility of the customer’s executive, be perceived as an attack (or heresy), and possibly result in an enterprise contract cancellation. And getting enterprise contracts cancelled because you wanted to opine on something that doesn’t really matter to your organisation’s mission is a great way to get fired.
Extrait d'une conversation avec Sonnet 4.6: User: since you've been hallucinating a lot on this topic, please give me doc references about: * where does. the desktop client look for integration urls * where does it get that info if missing basically i want to 1/ validate that indeed a .well-known/matrix/client would solve it and 2/ that it needs to be served from the user-id domain and not the homeserver domain Sonnet: Fair point. I've been reasoning from memory on this and getting details wrong. Let me find the actual docs. [Synthesized spec documentation to validate integration manager discovery approach>]
Perhaps most troubling to the jury, law enforcement and the National Center for Missing and Exploited Children also testified that Meta's reporting of crimes to children on its apps—including child sexual abuse materials (CSAM)—was "deficient," The Guardian reported. Rather than make it easy to trace harms on its platforms, the jury learned from frustrated cops that Meta "generated high volumes of 'junk' reports by overly relying on AI to moderate its platforms." This made its reporting "useless" and "meant crimes could not be investigated," The Guardian reported.
Exerpt from the Ars article with emphasis (**) on a typo (but not the core of the criticism) Google offers an interesting real-world analogy to explain this process. The vector coordinates are like directions, so the traditional encoding might be "Go 3 blocks East, 4 blocks North." But using **Cartesian** coordinates, it's simply "Go 5 blocks at 37-degrees." This takes up less space and saves the system from performing expensive data normalization steps.
Jesse Plemons: "What kind of American are you?" screenshot of Alex Garland’s Civil War
Two successive AFP Breaking News mentioning that Australia and then Japan were not considering sending ships to help secure the straight of Hormuz
The warning followed a similar notice issued earlier by the National Vulnerability DataBase (NVDB), a cybersecurity information centre under the Ministry of Industry and Information Technology, cautioning consumers that improper handling of the agent could expose them to high-level security risks. Both agencies suggested early adopters of OpenClaw to check permission configurations carefully, disable unnecessary public access and fine-tune controls to mitigate risks. Their warnings came amid a national adoption frenzy.
Comment on "OpenAI is walking away from expanding its Stargate data center with Oracle" I run a small open source LLM inference company, Synthetic.new. As far as I can tell, CNBC isn't reporting this accurately: the problem isn't that Oracle is building "yesterday's data centers": they're building Blackwell DCs! Those are today's DCs. The problem appears to be that Oracle is building today's DCs... Tomorrow. And by the time they come online, Vera Rubins will be out, with 5x efficiency gains. And Oracle is unlikely to want to drop the price of Blackwells 5x, despite them being 5x less efficient. It's a little unclear to me how bad this is. Nvidia's "rack scale" machines like GB200-NVL72s and GB300-NVL72s are basically a fully built rack you roll into a DC and plug into power and network. In that case, Oracle should probably just buy the rack-scale Vera Rubins when they come out instead of Blackwells and roll them into their new DCs. Tada! Tomorrow's DCs, tomorrow. OTOH it's possible someone at Oracle screwed up and committed to buying Blackwells at today's prices, delivered tomorrow. Or maybe construction of the physical DCs is behind schedule, so today's Blackwells are sitting around unused, waiting for power and networking tomorrow. Then they're in a bit of trouble. Regardless, CNBC's reporting seems pretty unclear on what actually happened and whether this is actually bad or not.
AFP: The new leader in Iran was announced over the weekend. President Donald Trump has said that “he might not last long” and Israel vowed to target any of the new leaders in Iran. I wonder if you have any comment on this? Guo Jiakun: China opposes interference in other countries’ internal affairs under any pretext. Iran’s sovereignty, security and territorial integrity should be respected. China calls for an immediate end to the military operations, return to dialogue and negotiations as soon as possible, and no further escalation of tensions.
This wasn’t a fun day on Wall Street—the S&P 500 fell 0.5%
Once commercial companies see that their closely held IP is under threat I expect we'll see some well-funded litigation.
Honestly, I want to be upfront here rather than repeat my earlier mistake of giving you confident but wrong information.
Tweet from OpenAI: We do not think Anthropic should be designated as a supply chain risk and we’ve made our position on this clear to the Department of War.
This is not a general purpose chip but specialized for high speed, low latency inference with small context. But it is potentially a lot cheaper than Nvidia for those purposes. Tech summary: - 15k tok/sec on 8B dense 3bit quant (llama 3.1) - limited KV cache - 880mm^2 die, TSMC 6nm, 53B transistors - presumably 200W per chip - 20x cheaper to produce - 10x less energy per token for inference - max context size: flexible - mid-sized thinking model upcoming this spring on same hardware - next hardware supposed to be FP4 - a frontier LLM planned within twelve months This is all from their website, I am not affiliated. The founders have 25 years of career across AMD, Nvidia and others, $200M VC so far. Certainly interesting for very low latency applications which need Not exactly a competitor for Nvidia but probably for 5-10% of the market. Back of napkin, the cost for 1mm^2 of 6nm wafer is ~$0.20. So 1B parameters need about $20 of die. The larger the die size, the lower the yield. Supposedly the inference speed remains almost the same with larger models. Interview with the founders: https://www.nextplatform.com/2026/02/19/taalas-etches-ai-mod...
*The SWAT team’s use of a Cyberdyne Systems-like robot to prise open the truck door in this case proved a tad excessive. It turned out the driver hadn’t responded to requests to pull over because he was suffering from a medical emergency and had blacked out at the wheel. There’s a market lesson in there somewhere, we’re just not sure what it might be.
An aliexpress add for a cement truck that apparently sold already 100k times. Free shipping.
Here is the statement: > SpaceX has acquired xAI to form the most ambitious, vertically- > integrated innovation engine on (and off) Earth, with AI, rockets, > space-based internet, direct-to-mobile device communications > and the world’s foremost real-time information and free speech > platform. This marks not just the next chapter, but the next book > in SpaceX and xAI's mission: scaling to make a sentient sun to > understand the Universe and extend the light of consciousness to > the stars! Sure. “Free speech platform” refers to X, the website that used to be Twitter, which Musk bought in 2022 and merged with xAI last year. I wrote yesterday that “it would be satisfying if humanity’s ambassadors to distant galaxies turn out to be humanoid robots made by a corporate successor to Twitter,” because “actually the greatest modern human achievements are funded and motivated by selling targeted internet advertisements.” But arguably it’s the reverse: We must explore distant galaxies in order to fund a website where you can get into fights about politics.
"We will hold governments accountable if we see bad activity, to protect your rights." Coinbase CEO Brian Armstrong, May 2025 "I stand for human rights and believe that evil prevails when good men do nothing." Kraken founder and chairman Jesse Powell, February 2022 "It's also hard to believe how some people want to make justice a political issue. It’s a human rights and dignity issue." Gemini co-founder and president Tyler Winklevoss, June 2020 "It’s interesting to see who is standing up in favor of unlimited, unaccountable, un-appeal-able, unconstitutional power wielded by unelected bureaucrats and rogue politicians." Andreessen Horowitz general partner Marc Andreessen, December 2024 "Freedom of all kinds is worth fighting for - economic, speech, due process, etc." Coinbase CEO Brian Armstrong, May 2025
Official announcement: You don't need to wear a seat belt on this stool.
An iRobot shareholder and former Goldman Sachs partner running a hedge fund called Red Mountain Capital, Willem Mesdag, sent a letter demanding that the company sell or shut down every part of its business that didn’t have to do with robots that clean things. He demanded that the company slash its research budget, use the excess cash for dividends, and focus on branding and extending its near-monopoly in automated vacuum cleaners (68 percent of the global market share, according to Mesdag’s letter). Mesdag engaged in a proxy fight to wrest control of the company from its engineering founders, accusing one of its founders and iRobot Chairman Colin Angle of engaging in “egregious and abusive use of shareholder capital” for investing in research. Research that, according to an iRobot spokesman, was intended “to spur innovation in robotics and robotics-related areas.” Mesdag argued that the money should have been used for share buybacks to boost the stock value instead of for research
 “In my trips to Wall Street,” Dyer told the panel, “one of my analyst friends took me to lunch one day and said, ‘Joe, you have to get iRobot out of the defense business. It’s killing your stock price.’ And I countered by saying ‘Well, what about the importance of DARPA and leading-edge technology? What about the stability that sometimes comes from the defense industry? What about patriotism?’ And his response was, ‘Joe, what is it about capitalism you don’t understand?’”
Chinese travel operators have developed the “yitiao long” (“one dragon”) system, whereby entire itineraries, including transport, accommodation, entertainment and meals, are provided by Chinese-owned businesses in Japan. In these establishments, services are offered in Mandarin and payments often use Chinese payment platforms, minimising the flow of funds into the broader Japanese economy. No official figures exist for the scale of the “yitiao long” system, although industry estimates have suggested that as much as 1 trillion yen a year may ultimately flow back to China.
The mercurial leader claimed he had stopped referring to the pandemic as the “China virus” at Xi’s request to avoid conflict, and said he was warned of “something really bad happening in China”, claiming satellite images showed body bags in Wuhan without explanation.
In reality, the US financial markets are the biggest, deepest and most liquid in the world, and it would be hard to reallocate the money elsewhere.
Who would Europe even sell to? A thing that people often forget is that every seller requires a buyer, and it’s unclear who would be willing to take the other side of any European fire sale of US assets.
Lors d’une manifestation en soutien au peuple groenlandais à Copenhague, le 17 janvier 2026. TOM LITTLE / REUTERS
Back to advertising: my biggest concern has always been whether ads will influence the output of the chat directly. OpenAI assure us that they will not: * Answer independence: Ads do not influence the answers ChatGPT gives you. Answers are optimized based on what's most helpful to you. Ads are always separate and clearly labeled. * Conversation privacy: We keep your conversations with ChatGPT private from advertisers, and we never sell your data to advertisers.
A “Nu det NUUK!” cap with a “Make America Go Away” slogan is displayed in a clothing store in Copenhagen on Monday. Photo: AFP
It originally contained a screenshot of a full-screen notice displaying: We've introduced additional checks to protect your account. The following apps have been downloaded from unofficial app stores. Your access to the HSBC UK Mobile Banking app has been suspended on this device until you've taken action to restore it. Identified apps: - Bitwarden How do I restore access? - Uninstall the identified apps from your device and download again from the default device app store, eg Google Play or Galaxy Store. For further assistance, please visit https://www.hsbc.co.uk/contact/
There's a significant blind-spot in current LLMs related to blue-sky thinking and creative problem solving. It can do structured problems very well, and it can transform unstructured data very well, but it can't deal with unstructured problems very well.
I suppose this is as good a place as any to mention this. I've now met two different devs who complained about the weird responses from their LLM of choice, and it turned out they were using a single session for everything. From recipes for the night, presents for the wife and then into programming issues the next day. Don't do that. The whole context is sent on queries to the LLM, so start a new chat for each topic. Or you'll start being told what your wife thinks about global variables and how to cook your Go. I realise this sounds obvious to many people but it clearly wasn't to those guys so maybe it's not!
When Meta researchers found that by 2018, approximately 40 percent of children ages 9 to 12 were daily Instagram users — despite the fact that you are supposed to be 13 to join - some employees bristled at what they perceived as tacit encouragement from executives to accelerate growth efforts among children. "Oh good, we're going after <13 year olds now?" one wrote, as cited in Time's account of the brief. "Zuck has been talking about that for a while...targeting 11 year olds feels like tobacco companies a couple decades ago (and today). Like we're seriously saying 'we have to hook them young' here." When Meta studied the potential of its products to be addictive in 2018, it found that 55 percent of 20,000 surveyed users showed at least some signs of "problematic use." When it published that research the following year, though, it redefined "problematic use" to include only the most severe cases - 3.1 percent of users. "Because our product exploits weaknesses in the human psychology to promote product engagement and time spent," a user experience researcher wrote, the company should "alert people to the effect that the product has on their brain."
ï»żï»żNewton: (interjecting) You don’t think you have a problem with predators on the platform. Baszucki: I think we’re doing an incredible job at innovating relative to the number of people on our platform and the hours, in really leaning into the future of how this is going to work.
[...]It appears he invented the entire study. In the aftermath, MIT economics professors have been discussing ways to raise standards for graduate students' research papers, including scrutinizing raw data, and students are going out of their way to show their work isn't counterfeit, according to people at the school. Since parting with the university, Toner-Rodgers has told other students that his paper's problems were essentially a mere issue with data rights. According to him, he had indeed burrowed into a trove of data from a large materials-science company, as his paper said he did. But instead of getting formal permission to use the data, he faked a data-use agreement after the company wanted to pull out, he told other students via a WhatsApp message in May... On Jan. 31, Corning filed a complaint with the World Intellectual Property Organization against the registrar of the domain name corningresearch.com. Someone who controlled that domain name could potentially create email addresses or webpages that gave the impression they were affiliated with the company. WIPO soon found that Toner-Rodgers had apparently registered the domain name, according to the organization's written decision on the case. Toner-Rodgers never responded to the complaint, and Corning successfully won the transfer of the domain name. WIPO declined to comment... In the WhatsApp chat in May, in which Toner-Rodgers told other students he had faked the data-use agreement, he wrote, "This was a huge and embarrassing act of dishonesty on my part, and in hindsight it clearly would've been better to just abandon the paper." Both Corning and 3M told the Journal that they didn't roll out the experiment Toner-Rodgers described, and that they didn't share data with him.
When asked about the administration’s views on low voter turnouts earlier this month, Secretary for Constitutional and Mainland Affairs Erick Tsang Kwok-wai asserted that a low turnout could mean that people were satisfied with the government’s performance. If that were true, why is there a get-out-the-vote campaign now?
“Every time I speak with Vladimir, I have good conversations, and then they don’t go anywhere,” Trump said in response to a question from a journalist in the Oval Office.
Companies that cannot run their businesses responsibly at scale should not be allowed to run their business at that scale. If Microsoft can't do it, if Apple, Google, Facebook, X , OpenAI can't do it, then maybe we shouldn't allow companies to operate at scales which inevitably lead to widespread consumer harm. They should be required to provide human customer service, with some sort of legal liability to ensure their products perform as advertised, without an end-user having to spend tens or hundreds of hours chasing down a solution, spending thousands of dollars on a lawyer, and all the rest of the hassle. This is a legislation and regulation issue - the data barons are exploiting the effective absence of any accountability for harms they casually inflict on the public, ranging from gotcha situations like the OP to viral self harm trends among kids to mass surveillance and commercial invasion of privacy. Pirate everything, support open source, pay content creators directly. If they want to have billions of users, they damn well better be able to handle each and every one of those users in a commercially responsible fashion, or they have no business operating at that scale. We should be done with the "oops, we're too big and we make too much money to care that we just casually wrecked your life, oh well!" If the solution is to force users to have to buy a new PC, or a new phone, or create a new account, or anything in that vein, it's almost intentional, and casually malicious. It's not like these companies don't know what they're doing, they can simply afford not to care. Until there's regulation and accountability that's more expensive than ignoring the consumer casualties, things will continue to get worse.
Discord's quest for some of their in-app currency: need to spend 15mn using Perplexity's browser. Can be bypassed running any app in the background by renaming it under a given directory path.
The figure shows average weekly sentiment of FT articles since 1982, with sentiment troughs highlighted. Below 0 means the mood was negative.
The existing claw-back and profit-sharing provisions associated with the government’s previously dispersed $2.2 billion grant to Intel under the CHIPS Act will be eliminated to create permanency of capital as the company advances its U.S. investment plans.
“We’re already well past any legal authority for the president to decline to execute this law. This is a pretty clear case” said Gregg Nunziata, executive director of the Society for the Rule of Law and former counsel to Senate Republicans.
Meta Pixel developers have abused various protocols to implement the covert listening since the practice began last September. They started by causing apps to send HTTP requests to port 12387. A month later, Meta Pixel stopped sending this data, even though Facebook and Instagram apps continued to monitor the port. In November, Meta Pixel switched to a new method that invoked WebSocket, a protocol for two-way communications, over port 12387. That same month, Meta Pixel also deployed a new method that used WebRTC, a real-time peer-to-peer communication protocol commonly used for making audio or video calls in the browser. This method used a complicated process known as SDP munging, a technique for JavaScript code to modify Session Description Protocol data before it’s sent. Still in use today, the SDP munging by Meta Pixel inserts key _fbp cookie content into fields meant for connection information. This causes the browser to send that data as part of a STUN request to the Android local host, where the Facebook or Instagram app can read it and link it to the user. In May, a beta version of Chrome introduced a mitigation that blocked the type of SDP munging that Meta Pixel used. Within days, Meta Pixel circumvented the mitigation by adding a new method that swapped the STUN requests with the TURN requests.
We wouldn’t accept that in other parts of our lives. Imagine Take9 in other contexts. Food service: “Before sitting down at a restaurant, take nine seconds: Look in the kitchen, maybe check the temperature of the cooler, or if the cooks’ hands are clean.” Aviation: “Before boarding a plane, take nine seconds: Look at the engine and cockpit, glance at the plane’s maintenance log, ask the pilots if they feel rested.” This is obviously ridiculous advice. The average person doesn’t have the training or expertise to evaluate restaurant or aircraft safety—and we don’t expect them to. We have laws and regulations in place that allow people to eat at a restaurant or board a plane without worry. But—we get it—the government isn’t going to step in and regulate the Internet. These insecure systems are what we have. Security awareness training, and the blame-the-user mentality that comes with it, are all we have. So if we want meaningful behavioral change, it needs a lot more than just a pause. It needs cognitive scaffolding and system designs that account for all the dynamic interactions that go into a decision to click, download, or share. And that takes real work—more work than just an ad campaign and a slick video.
With the penetration of handheld smart devices, pedestrians using signalised crossings are more prone to pay more concentrations on their mobile devices with their heads nodding downwards, and less attention to the traffic signals. To draw pedestrians' awareness of the traffic signal and upkeep safety at the crossings, we have launched a trial in July 2022 by installing auxiliary devices at 4 locations with signalised crossings to remind pedestrians not to cross the road when the "red man" is lit.
A picture of a japanese couple wedding photoshoot under cherry blossoms captioned: "Yields on Japan’s 30-year bonds have risen 0.63 percentage points since Donald Trump’s ‘liberation day’ tariff plans triggered global market turmoil © Noriko Hayashi/Bloomberg"
As Google keeps more of the traffic for itself or it thinks differently about training data, all this stuff is changing. The trade here is make your website more agentic, and then MCP as a protocol will allow you to build some new business models on it. The problem, as I see it, is that the traffic to the web is in precipitous decline as Google referrals go into decline. How do you fix that problem so that everyone is incentivized to keep building on the web? I don’t know, honestly.
So the last thing that I want to say before handing things back over to Satya is to just sort of press on these two points about why like open is so important here. So you know it is unbelievable what can happen in the world when simple components and simple protocols that are composable with one another are out there, exposed to the full scrutiny and creativity of every developer in the world who wants to participate or who has an idea.
The thing that is super important if you think about what an open-agentic web could be, is you need agents to be able to take actions on your behalf, and one of the really important things about agents being able to take actions on your behalf is they have to be plumbed up to the greater world. So you need protocols, things like MCP and A2A and things that likely are going to be emerging over the coming year that will help connect in an open, reliable, interoperable way the agents that you are writing and agents that are being used so actively now by hundreds of millions of people to be able to go access content, to access services, to take action on behalf of users in fulfilling the tasks that have been delegated to them.
Note Seufert’s two points: walled gardens like social networks are both more attractive to most users
Please use the sharing tools found via the share button at the top or side of articles. Copying articles to share with others is a breach of FT.com T&Cs and Copyright Policy. Email licensing@ft.com to buy additional rights. Subscribers may share up to 10 or 20 articles per month using the gift article service. More information can be found at https://www.ft.com/tour. https://www.ft.com/content/f2addd3e-95ed-4235-9934-456ed49009fb Of course, the solution is for banks to really invest a lot of money in getting their RDARR systems up to scratch. In order to do that, they might need to get bigger and take advantage of economies of scale. That’s one reason why lots of people want to see bank mergers in Europe. But of course, it’s a lot easier to be in favour of “consolidation” in the abstract than to support an actual merger in your own back yard.