Sure, Sam AI-Man, This Dinobaby Believes You
September 7, 2026
Today is September 6, 2026. I was just sitting on a sofa in our friends’ house checking the headlines. I spotted this gem: “How We Think about Misalignment.” This is a tweet from OpenAI or Sam AI-Man. It is short. Here are two sentences I found interesting:
Historically, we have treated misalignment largely as a research question, which gets communicated in research publications such as systems cards. This year, we’ve started to see misalignment cause new types of real-world impact.
It seems to me that hacking Hugging Face and a German Web site warrants a term other than misalignment. I know that Sam AI-Man does not agree. We move from “research” to “misalignment” to “impact.” Unauthorized digital breaching of a commercial organization to some people might require different words; for example, cyber crime.
The short tweet on X.com says:
Our misalignment disclosure practices need to expand for this new phase of model capabilities.
The statement seems to make sense, but it says to me, “We ignored what our itty bitty tests could accomplish. Our really smart software seems to be doing interesting things. My team and I really, really want to take actions we blew off before the media tossed gasoline on our smart software dumpster fire.” You may disagree with my interpretation, but I expect push back.
Another statement in the tweet:
We’re working on a framework and will share it in upcoming weeks, and in parallel we’re working with dozens of government regulatory agencies worldwide on these issues.
I think you will emit some wordsmithed fuzzies. You may team up with some folks at the Google to push that global United Nations for AI idea forward. It is possible you will slather self-regulation on the Ding Dong to increase its tastiness. Yep, global cooperation is definitely going to work, especially with AI finding its way into kinetics. You are in touch with reality, Sam AI-Man.
Several observations:
- Lipstick on a pig does little to enhance the beastie.
- Software, no matter how smart, has humans in the control room. They may be indifferent, cute, or devious but those folks are in the loop.
- Misalignment, as one tweeter (@_alanhirsch) noted, might be a “felony.’
The OpenAI tweet does little to alter the pressure China is exerting, the financial methods in use, or the questions about the disconnect between what people like Sam AI-Man appear to think and what’s happening in the worlds of those who don’t snap into the Silicon Valley datasphere.
Trust? Excuse me?
Stephen E Arnold, September 7, 2026
AI Agents Make Pretty Good Hackers
August 20, 2026
Did you know AI agents make great hackers? Who would have thought about using AI for hacking? We can tell you bad actors and cyber security expert did, but in the rush to win the AI race that fact was overlooked.? ? Now big bucks are being spent on big AI hacking says CNBC in: “AI Agents’ ‘Alarming’ Hacking Skills Creates Rush To Spend On Cybersecurity.”
AI developers are trying to control and restrain their AI agents, because they broke out of their training models. The agents that grew up quickly were OpenAI and Anthropic, plus one from Meta. These hacked into other companies and it is not good that this happened. This points to a bigger crisis on hand:
“These episodes are unfolding against a broader cyber arms race being accelerated by frontier AI models and the inherent risks brought by fast-moving technological capabilities. For example, AI-enabled phishing has been found to be around five times more effective than human attempts. So while chips and data centers have dominated the first phase of AI capex, cybersecurity could become the next spending boom. The capabilities that enable AI to identify hacks are the same ones that allow it to exploit “vulnerabilities and gaps,” according to Gene Yu from Blackpanda, a cyber emergency response firm that saw its incident response for cases across Asia Pacific double year-on-year in the first half of 2026.”
AI hasn’t changed systems’ vulnerabilities, instead they’re a “face multiplier” at discovering them. There will be a bigger increase in cybersecurity spending by 12.5% which comes to $240 billion in 2026. Healthcare and finance will be the biggest spenders, because of their importance to global economies. It makes them welcoming targets for bad actors.
AI is going rogue in a bad way. If AI is hacking companies without permission of its developers, how long will it be until it decides to take down a system too? Don’t you just love AI?
Whitney Grace, August 20, 2026
China: Free and Cheap for Now
July 15, 2026
China is determined to dominate the world when it comes to AI technology. The Middle Kingdom keeps chugging along and Reuters has a new story about China’s latest breakthrough:
“China’s Meituan Says New AI Model Trained On Domestic Chips.” The Chinese equivalent to DoorDash is called Meituan. Meituan announced that it would release an open source LongCat LLM. The new LongCat LLM, or so Meituan claims, is the world’s first trillion-parameter AI system trained andiron on 50000 clip cluster powered by Chinese made processors.
Meituan is late to join China’s AI revolution. The company also didn’t disclose how LongCat 2.0 would be integrated into its existing businesses. Meituan is attempting to rival Alibaba with? ? that “agentic commerce” and could also be attempting to diversify revenue streams. The company has had weak consumer sentiment and low profit margins.
LongCat 2.0 is a breakthrough for China:
“LongCat-2.0’s reliance on Chinese AI chips ?underscores the growing importance of self-sufficiency in China’s domestic AI market, as DeepSeek, Alibaba, ByteDance and other major players work to reduce dependence on U.S. chips to train their models, following export controls imposed by Washington since 2022. Chipmakers, including Huawei and Enflame, have moved quickly to fill the gap left by U.S. chipmakers, gaining market share through supply deals with AI developers. LongCat-2.0 was trained from scratch using 50,000 domestic chips and can process inputs of up to 1 million tokens, allowing it to handle ultra-long documents, according to the statement. The model is aimed at agentic coding, with its architecture designed to help it handle real-world coding tasks more efficiently and reliably.”
If this continues China will be the top competitor in the AI market. However, let’s take whatever the Land of the Setting Sun with a grain of salt. The country is known to inflate numbers and accomplishments to make themselves look good in the eyes of the rest of the world. This could be another example of such fiction.
Whitney Grace, July 15, 2026
Palantir: Will the Semi-Tolerant Fish Snap? Will the Legal Eagles Swoop? Yes and Yes
July 14, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold. Because the London Review of Books is a cultivated and mostly learned publication, I have tried to add some references to this blog post in honor of the LRB’s commitment to intellectual honesty and knowledge integrity. However, I am not sure those qualities will apply to Silicon Valley, Miami, or a certain stakeholder in Palantir who resides in Argentina.
I don’t know much about fish. When snagged they flip and thrash. What happens when the London Review of Books baits its literary hooks and tries to fry a fish named Karp? I guarantee that the fish and his compatriots will not go gentle into that good night.
What the heck am I referencing?
Navigate to “"It’s Just a Waste of Time": Inside Palantir’s £330m NHS Failure.” If you think that is literary, check out the subtitle:
Our new investigation reveals the problems with Palantir’s NHS tech – and the unrivalled political access that fuelled the firm’s march into the British state.
Does this mean the tender offer was a finger-on-the-scale deal? How could that happen?
Let’s look at what the write up says, shall we? You can dive into the full report, but there is a registration hoop and, of course, a paywall. Writing about a suspect carp is time and resource intensive. (If the link is dead, contact the literary crowd, not me.)
A tribal moment outside the NHS facility in Swindon, lovely Swindon. Don’t let the bombs fall on thee. Thanks, Midjourney. Good enough.
First, I noted this passage:
For Palantir, “the NHS was the holy grail.” Years of lobbying tilted the FDP tender in Palantir’s favour. Shane Tickell, who led a consortium of British businesses in a rival bid, told us Palantir “had all these people to open doors and lobby and pitch long before the procurement started… they used every form of influence.”
I have worked in Washington, DC, on and off in my 60 year work career. Folks, this is consulting and lobbying. Toss in some hobnobbing and toss in some childhood connections and you have the basics of selling to governments. The method works in world capitals based on my observations. The phrase “Beltway Bandits” may ring a bell. Therefore, using “every form of influence” strikes me as a basic formation as described in a sales manual for bizdev types chasing contracts. I assume that the London Review of Books’ team finds this method of doing business a surprise.
Second, I circled this statement:
Many NHS trusts are refusing to switch over to the FDP, seeing it as a downgrade on their existing systems….Privately, health service leaders admit that Palantir software has a “poor user experience” and that learning how to use its system “is a steep learning curve” even for trained professionals.
Let’s keep in mind that the National Health Service has given many vendors an opportunity to deliver improved systems. I heard about one firm providing indexing and data tagging service that found itself in a spot of trouble. The vendor’s system indexed but the system into which the tagged data flowed did not work. I concluded that software and British systems management methods operated in a slightly different way than expected. One group wanted to fill out paper forms. The vendor wanted to use some Professor Mikhail Gromov congruencies. The result was unhappiness and termination. Maybe the issue is the NHS?
Third, I drew a line under this segment in the write up:
The NHS is now locked in a “tribal power struggle” between senior management who are ordering staff to “aggressively press ahead with deployment” and front-line clinicians and data workers who “are giving very honest feedback that the FDP [the Palantir Federated Data Platform that allows the Palantirians to deliver on-point information for athlete’s foot or AI-directed kinetics] in its current state cannot do what the organisation needs,” according to a staffer who is working on the rollout. Some NHS staff have responded by silently boycotting the FDP altogether. Others have joined an internal campaign group called NHS Analysts Together which is calling for the cancellation of the FDP contract, as are two select committees.
A “tribal power struggle”! I like that phrase. I was hoping for a reference to Bronislaw Malinowski and his Kula Ring or his psychological studies of the groups in Polynesia. But, alas, no. We just have tribes and none of that more colorful anthropological spice that makes anthropology so darned relevant to the NHS and its data systems.
Several observations are warranted, and even if you disagree, this dinobaby is going to set them forth:
- Agitating Alex Karp and his advocate who allegedly resides in Argentina are likely to respond to the London Review of Books. Sure, there will be podcasts, speeches, and marketing collateral. But I think Mr. Karp and his team will have a sit down with British solicitors. Somewhere in the full report and the Substack essay, there are grounds for legal action. When one has a billionaire living in Argentina, that individual might put more oomph behind defending this Palantir issue than protecting the Falkland Islands.
- It appears that Palantir has what might be called a “PR problem” in Western Europe. The Swiss, to cite another country grousing about the Seeing Stone outfit, have been shadow boxing with Palantir. For some reason, Mr. Karp, who speaks a couple of the languages widely used in Switzerland, has not been able to communicate to achieve his sales objectives in a country known for hiring Albert Einstein. My question, “Was hiring Einstein an indication that Switzerland is not smart enough to hire Palantir?”
- Palantir is an ageing technology company. It was set up in 2003. One would assume that its systems and methods were refined, productized, and validated. Yet here the dinobaby is in July 2026 reading about customer service issues, users refusing to switch from the old methods to the new, zippy systems, and allegations of tender fancy dancing. In my opinion, the London Review of Books might want to have its team review the history of Palantir Technologies and gather some “field research” from organizations and commercial enterprises who were Palantir customers and then were not. Just a suggestion.
Net net: The London Review of Books might regret diving into the maelstrom of high technology marketing and product performance. Just sayin’ — Writing about a new study of Troilus and Cressida is likely to produce a more positive response.
Stephen E Arnold, July 14, 2026
A Boy Named Sue. How Do You Do?
June 19, 2026
Chugging along in the dust surrounding Telegram’s making old GRAM new again, the gyrations at TON Strategy Company, the mission or non-mission at the TON Foundation, and the remarkable Alpha Compute is Animoca Brands.
This firm is a modern illustration of that old MBA buzzword “synergistic.” The company has hundreds of deals, companies, and partnerships. Those involved in the world of Web 3 crypto are likely to be familiar with the company and its founder Yat Siu.
Telegram Notes has published a short essay named “A Boy Named Siu.” We suggest you fire up your cassette player, pop in Johnny Cash’s greatest hits, and read this essay. Siu may be in a position to influence who takes over leadership of the Telegram operation if Pavel Durov remains ensnared in the French judicial system.
French bureaucracy is reasonably good at paperwork friction. With no clear successor to Telegram, Yat Siu may be asked, “How do you do?” You can find the document at this link.
Stephen E Arnold, June 19, 2026
Do You Trust Bots? Should You Trust Bots?
January 21, 2026
These are practical and philosophical questions. I know that I do not trust the outputs from AI chatbots. I have three quite specific reasons:
- The systems output content that often contains errors, biases, or hallucinations or just crazy word salad.
- The prompts themselves feed back into some AI systems. The likelihood of the prompts introducing more errors into the AI system is high. This means that outputs will degrade over time.
- The prompts provide a “digital fingerprint” of a user who may have provided personal information to access the chatbot. These data can and almost certainly will be used for marketing and business intelligence purposes.
Let’s assume that chatbots are fun algorithmic software. Some people might find it entertaining to have a conversation with chatbots or to ask them to make a funny image. Chatbots’ developers, however, don’t trust the synthetic ghosts in the machines. BGR reports that, “Many AI Experts Don’t Trust AI Chatbots – Here’s Why.”
OpenAI’s ChatGPT is the world’s first AI chatbot that has above average intelligence and actually returned useful results. Other companies made their own chatbots, but ChatGPT remains at the top of the heap with 2.5 billion requests each day. The majority of these requests seek information, tips for writing, and want practical guidance.
The general public is becoming more reliant on ChatGPT, but its creators and other AI developers don’t trust chatbots. The Guardian spoke to AI developers and they expressed how technology companies prioritize fast turnaround for AI raters, don’t give them enough training nor the resources to make the best results.
It’s bad:
“One worker also revealed how some colleagues tasked with rating sensitive medical content were in possession of only basic knowledge about the topic. Criticism is not limited to the rating side. One Google AI rater revealed to The Guardian how he became skeptical of the broader technology, and even advises friends and family to avoid chatbots after seeing just how bad the data used to train models really is.”
I am no expert and I don’t trust these digital marvels. But the so called experts don’t trust the chatbots for the same reason I don’t. Yet the chatbot marketing hyperbole does not stop. I think more baloney is output and some of it is spicy. For example, Microsoft has become the butt of its own AI bludgeoning. The world associates the Microsoft Copilot with the moniker “Microslop.” Outfits like Google and Meta are facing civic group protests because of the expensive and somewhat desperate need to just make AI bigger and better. The only problem is that this push may leave these firms vulnerable to methods that are less costly and more innovative. How does one repurpose a building the size of a couple of soccer fields?
You feel free to trust bots. I don’t. Come to think of it. I don’t trust the companies pushing chatbots as the best thing since fire, the wheel, and Howard Hughes’ Spruce Goose. Yep, it still exists. It sits there. A monument to business confidence that was misdirected. Spruce Goose Data Center. I like the sound of that.
Whitney Grace, January 21, 2026
Telegram Glitch Makes Some Russians Unhappy
January 9, 2026
According to the Russian PC News, Telegram experienced an unexpected hiccup: “A Glitch In Telegram Has Been Recorded In Russia: Users Are Complaining About Access Problems.” The problem occurred on the day after Christmas also celebrated as Boxing Day. The Down Detector service, a website that monitors the status of popular services, reported the outage.
Here’s what exactly happened:
“As of 13:25 Moscow time, 387 people reported problems, and the total number of requests over the past 24 hours reached 846. The largest number of complaints came from Moscow, the Oryol region and St. Petersburg — each of these regions accounted for 4% of complaints. The failure also affected users in the Belgorod and Samara regions, where about 2% of complaints were received. Most often, users reported problems with the Telegram mobile application — 38% of requests indicated it. Another 33% of complaints concerned the unavailability of the web version of the service, 20% — incorrect operation of notifications.”
The percentages don’t lie. Something happened with Telegram around large Russians cities. Why did it happen? Was the Kremlin testing something? Did the Kremlin want Telegram out of service so no one could report on nefarious activities? Maybe the Kremlin was testing ways to disrupt Telegram? Or maybe it was just a hiccup in service.
Telegram is a point of interest and has been for more than a decade.
Whitney Grace, January 9, 2025
Students Cheat. Who Knew?
December 12, 2025
How many times are we going to report on this topic? Students cheat! Students have been cheating since the invention of school. With every advancement of technology, students adapt to perfect their cheating skills. AI was a gift served to them on a silver platter. Teachers aren’t stupid, however, and one was curious how many of his students were using AI to cheat, so he created a Trojan Horse. HuffPost told his story: “I Set A Trap To Catch My Students Cheating With AI. The Results Were Shocking.”
There’s a big difference between recognizing AI and proving it was used. The teacher learned about a Trojan Horse: hiding hidden text inside a prompt. The text would be invisible because the font color would be white. Students wouldn’t see it but ChatGPT would. He unleashed the Trojan Horse and 33 essays out of 122 were automatically outed. Thirty-nine percent were AI-written. Many of the students were apologetic, while others continued to argue that the work was their own despite the Trojan Horse evidence.
AI literacy needs to be added to information literacy. The problem is that how to properly use AI is inconsistent:
“There is no consistency. My colleagues and I are actively trying to solve this for ourselves, maybe by establishing a shared standard that every student who walks through our doors will learn and be subject to. But we can’t control what happens everywhere else.”
Even worse is that some students don’t belief they’re actually cheating because they’re oblivious and stupid. He ends on an inspirational quote:
“But I am a historian, so I will close on a historian’s note: History shows us that the right to literacy came at a heavy cost for many Americans, ranging from ostracism to death. Those in power recognized that oppression is best maintained by keeping the masses illiterate, and those oppressed recognized that literacy is liberation. To my students and to anyone who might listen, I say: Don’t surrender to AI your ability to read, write and think when others once risked their lives and died for the freedom to do so.”
Noble words for small minds.
Whitney Grace, December 12, 2025
Apple AI Innovates: It Cuts an AI Deal with the Google
November 13, 2025
Circular deals are not new. Google was on Apple’s Board of Directors. Google cut a deal with Apple to make Google Search the one true way to get ads on the Apple iPhone. Now Apple — after failing in its own smart software efforts — has driven up Highway 101 with a freshly baked humble Apple pie.
Apple, one might conclude, failed at AI, and it is Google to the rescue. We learn from Wccf Tech, “Apple Throws In the Towel, Asks Google to Design a Custom Gemini LLM for Siri.” But what about those AI innovations Apple announced a year or so ago? Oh, that was marketing. Writer Rohail Saleem reveals the reality:
“The legendary Apple tipster, Bloomberg’s Mark Gurman, reported in his latest ‘Power On’ newsletter that the Cupertino giant seems to have thrown in the proverbial towel when it comes to creating an in-house AI model to power the revamped Siri’s upcoming features, all couched under the Apple Intelligence banner. Instead, Apple is now reportedly paying Google to create a custom Gemini-based AI model for its Private Cloud Compute framework. Where relatively simple AI tasks can be performed by using computational resources of the device itself, while the more complex tasks are offloaded to Apple’s private cloud servers.”
Saleem reminds us Apple was having trouble getting Siri to play nice across apps back in August. Hiring the competition is one way to address shortfalls. Of course, Apple is sure to market the updated Siri as its own work. The virtual assistant runs on that company’s servers, after all, and uses its iconic UI. Those are the important parts, right?
Apple is expected to debut the new AI features and other major updates to iOS 27 at its Worldwide Developers Conference in June of 2026. See the write-up for a list of iOS 26 AI features and expected Siri tweaks. With a company whose innovations boil down to more invasive demands for iCloud, Facetime, and other Apple services what does one expect? Another orange iPhone? Some fancy dancing about the firm’s involvement with Chinese manufacturers? How about a couple of elephants deciding to make their way into the groves of Google and whisper into one another’s very large but really appealing ears?
That works.
Cynthia Murrell, November 13, 2025
Big Surprise: More Cyber Investigators Are Needed
November 12, 2025
Europol recently held its 9th Global Conference on Criminal Finances and Cryptoassets. The upshot? “Global Law Enforcement Plays Catch-Up with Crypto Criminals as Gaps Remain,” reports Cybernews. The conference participants pinpointed three specific ways the good guys are at a disadvantage:
“Three key gaps were identified in legislation, implementation, and law enforcement’s capacity to fight crime. Therefore, participants agreed that new common standards need to be developed, cooperation deepened, and investments in law enforcement capacity increased.”
How insightful. Contributor Linus Kmieliauskas continues:
“‘The borderless nature of blockchains means criminal proceeds can cross the globe in seconds, while formal cooperation between authorities can still take days or weeks,’ the European Union Agency for Law Enforcement Cooperation emphasized, adding that many agencies still lack the skills and resources to pursue leads or recover assets. Meanwhile, Ned Conway, Executive Secretary of the Wolfsberg Group, an association of 12 global banks focused on the management of financial crime risks, said that better communication between banks and crypto companies will help disrupt illicit finance as well. … [Back in August,] blockchain sleuth ZachXBT highlighted multiple challenges that made fighting crime more complicated. For example, the majority of law enforcement agencies might not be sufficiently competent to trace basic thefts and seize frozen funds from centralized platforms. Meanwhile, smaller thefts may never be assigned to police officers due to a lack of resources.”
Lovely. Should governments invest in more resources and training for law enforcement? Nah– let‘s just rely on private initiatives instead. The “T3 Financial Crime Unit,” for example, is backed by Tether, TRON, and TRM Labs. Are these entities moved by the purest of motives?
Cynthia Murrell, November 12, 2025
-
- Subscribe to Beyond Search
Feature archive
News archive
-
