Beyond Search

South Korea: Not Just Smart Hyundai Venues. It Is Also AI Censoring and Pre-Censoring

2026-06-05T14:01:21Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

I drive a super sexy blue Hyundai Venue. When I drive by an old age horn, the babes wave at me from their walkers and wheelchairs. I know I am cool. My Venue is smart. It beeps when I am in a parking space. It beeps and jerks the wheel when I am navigating roads under construction. Once in a while, the car beeps, flashes a red light, and automatically applies the brakes when I am creeping along in a traffic jam. Yep, South Korea has that smart software nailed.

Pre-censorship and real-time censorship appear to deliver some surprises to a small online publishing firm in South Korea. But AI is good enough, right MidJourney?

The country is now pioneering in a field translated for me by a “free” online translation service: Pre-censoring. I read about this nifty smart software sitting in my Venue as the beeps warned me of some impending disaster. The article is titled in this crystal clear way: “Even If You Upload an Image, Is It Pre-Censored? The Background of the Confrontation of Pros and Cons.” According to the free translation service the write contains this statement, and I quote:

The policy of pre-censoring images posted to the domestic Internet community with artificial intelligence (AI) has also become visible. Public opinion is divided between the need for a social safety net and the violation of freedom of expression.
The amendment to the Telecommunications Business Act, called the Prevention of N-Bang Act, came into effect in 2021. The structure monitors and responds in advance to content uploaded to SNS, messengers, and communities using AI. Until now, it had been limited to video files, but confusion soon grew as it became known that image files were also included in management.

I like the “N-Bang” bound phrase. Usually rules and regulations are less… suggestive. Confused. I poked around and located this article: “South Korean Online Communities Will Need to Scan Every Images with AI Censorship Tools.” The main idea seems to me, if the translation is sort of correct, is:

Due to recent regulation changes… the South Korean government is requiring internet communities and forum owners to scan every user uploaded images and videos on their website, by AI. The hardware to run these AI models are also not provided by government, website owners have to buy datacenter grade Nvidia GPUs by themselves, putting financial pressure to small businesses and forums. Websites will need to implement these hardware and software features, starting immediately from July 1st, [2026]

Several observations seem warranted:

Will the smart software perform in a manner similar to that in my Hyundai Venue: False beeps, erroneous beeps, and beeps from out of nowhere? (Hey, I’m parked with the motor running, and my Venue just beeped. Because the beeps are the same frequency, I am not sure what the problem is. I will lock my doors.
AI systems appear to have a few issues; for example, the systems hallucinate. Has South Korea figured out how to make smart software not output erroneous information; for example, an image posted on social media of a father splashing in a pool with his two young children? I am confident that some hallucinations will occur; for example, child cruelty, attempted murder by drowning, an image destined for a CSAM site on a Dark Web service, etc. But I assume South Korea’s AI does not have this problem.
The pre-processing and the real-time processing computational loads are zero problemo for those in the online delivery chain. We checked a single image online using five “smart” image identification services. It took about 15 minutes to get results in our “image horserace.” I assume that South Korea has engineered a workflow that does add time and cost to an online service that includes images.

Net net: I think the idea of pre- and real-time image filtering is interesting. Zipping through still images, video files, and any other included file type is no problem. Hey, now I am backing out of my parking space. My Hyundai Venue is beeping with false positives. The dog park is empty now, but the Venue is smart. It is protecting me from … something.

Stephen E Arnold, June 8, 2026

Google: Flips on the Yellow Alert Siren as Insecurity Spills across the Information Superhighway

2026-06-05T14:02:33Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

I spotted a news story on CNBC. The headline: “Alphabet Plans to Raise $80 Billion from Stock Sales to Fund AI Build-Out.” The report says:

In April [2026], Alphabet updated its full-year capital expenditure range to as much as $190 billion.

That’s a big number and typical of a Silicon Valley outfit struggling with the fear of being left behind or not invited to join a super elite coffee klatch at Stanford University.

Part of the problem is revealed unwittingly in a write up authored by the Google I/O team titled “All the News from the Google I/O 2026 Developer Keynote.” [I love those categorical affirmatives. Really “all”? Okay, why not just put the full text of the keynote in the developer blog? I noted several “in the white spaces” signals.]

A blast from an MBA class in the 1980s. The buggy whip case. Thanks, MidJourney, but the anxiety is just not hitting me.

Let me review what I noticed.

First, Google’s spate of announcements appears to say, “Hey, we are building a new type of operating system. Don’t worry. It will be great. But where is the killer announcement? This is a pile of “upgrades.” I was not fooled, and I know because Google is not suggesting it will spend almost $200 billion in the next six months. Yeah, that’s not a muscle twitch. This is a convulsion caused by anxiety sparked by insecurity.

Second, the change is search has benefited DuckDuckGo and some other Web search systems. Ecosia loved the Google announcements. Swisscows are mooing happily as well. Why not? Google effectively split those looking for information into those who want relevance ranked lists of links and those who want Google to tell them what’s the answer to their probably malformed prompt. Nice work, Google. You accomplished what none of these firms could achieve on their own: A spike in users. Kudos. Are you looking at the same Information Highway I watch?

Third, advertisers remain in a state of uncertainty. Their anxiety is equivalent to what the Google leadership lives with 24×7. Advertisers are losing control of how much they spend, where the ads appear, and useful information about what caused an order to arrive. One issue is transparency. Who knows if the AI goodies allow Google to steer revenue and clicks to Google’s benefit, not the advertisers’ benefit?

Adding to the concerns of management is the shingles some of Google wizards manifest. The Irish News via Bing published an interesting story; to wit: “Google Agrees to Talks with UK DeepMind Staff over Calls to Unionize.” My understanding has been that Google does what it wants, eschews traditional personnel methods, and does not want its wizards to unionize. The write up asserted:

Google has agreed to negotiate with staff at its artificial intelligence (AI) research lab over calls for a workplace trade union amid concerns about their work being used in the development of weapons.

Formal discussions will begin after Google rejected a request for union recognition for UK-based Google DeepMind workers. Members of its management team will now enter into negotiations with workers’ representatives via mediation body the Advisory, Conciliation and Arbitration Service (Acas). It is expected that discussions will lead to a formal ballot later this year for employees to vote over whether to unionize. If it went ahead it would mark a first for Google, which does not currently have a recognized trade union within the UK business or at DeepMind.

I think this is another signal that Google is nervous about how some of its AI experts view the firm’s policies. Plus, outfits like Anthropic, Grok, OpenAI, and who knows how many Chinese-linked firms are eager to hire those disaffected with the online advertising sales company.

Google spins money. However, one may now legitimately ask this question, “What if Google’s 25 years of unfettered revenue generation and power expansion is facing its buggy whip moment?” Can the firm make seat covers for the new fangled vehicles zipping down that Information Highway?

Stephen E Arnold, June 8, 2026

AI: Automating Dysfunction One Company At A Time

2026-06-06T20:35:02Z

If your CEO team doesn’t agree only anything, an AI system with a hefty price tag isn’t going to fix there situation says Stephen Smith in his article, “Your AI Strategy Is Just Your Dysfunction At 700 Horsepower.” Chief operators believe that spending wads of cash on expensive AI will solve all their problems. According to Gartner, it’s another story:

“In October 2025, Gartner published a survey of 413 marketing technology leaders running AI agents in pilot or production. Forty-five percent reported that the vendor-supplied agents weren’t delivering the business results the vendor had promised. About half admitted their data and technical stacks weren’t ready to run the agents at all. Half also said they didn’t have the technical talent to operate them. Gartner’s separate prediction puts more than forty percent of agentic AI projects on track to be cancelled by the end of 2027.”

Smith uses a horsepower metaphor to describe AI: “AI is horsepower. RevOps is the chassis, steering, brakes, and instrumentation. More horsepower on a broken chassis doesn’t create speed. It creates instability.”

The robots are not amused.

These companies are using AI to replace humans for “efficiency.” But “the next big thing” makes mistakes. Humans have to figure out the screw up and fix it. What does Smith suggest business leaders do? He recommends that they block out a section of time, make decisions, audit, and make fixing the mistakes one’s real job.

But what if the human with the skill to remediate has been terminated? Yo, Big Dog, just rev your engine.

Whitney Grace, June 8, 2026

Google and Personnel Management: New Taco Truck Operators Available

2026-05-31T15:13:10Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

I located another great example of the failure of modern hiring practices. The source of this information is the Alphabet Workers Union. I admit that it is a biased source. Those who work at Google are, by definition, the world’s smartest and most capable humans on planet earth. These humans create the must-have services that beg for personally identifiable information and do their level best to extract essential signals from clicks on their assorted online properties.

I read “We Are a Fighting Union.” I am not sure if the information in the Web page that displays is a call to action, a bunch of misfits complaining about work, or a disinformation operation powered by a Gemini agentic bot chain. Frankly I don’t care. What interests me is that the existence of “We Are a Fighting Union” provides a glimpse of the consequence of what might be called “flawed human resource intake, vetting, and training programs.” In a nutshell, the Web page illustrates the real or attributed behaviors of Google leadership. (Of course, I know the company is Alphabet, but, hey, I coined the word Googzilla for my monograph published by the now-defunct Infonortics Ltd. in the UK. Therefore, I stick with “Google” as an umbrella term and use “Googzilla” to signal brilliant actions taken by the company’s leadership.

I assume Google personnel has larger images of these individuals. I assume Google’s Lens technology or a variant can identify each individual as well. This snap from the AWU Web site suggests some Googlers are not happy campers:

What is the “problem” the union wishes to address. It is not working conditions. The offices, if one actually shows up, are comfy and provide some essentials like bean bag chairs, approved beverages, and some okayed snacks. Based on my scanning of the “We Are a Fighting Union,” the problem seems to be Google’s management failings. If Google were doing a good job, there would be no need to create a union. The AWU says:

Our union strives to protect Alphabet workers, our users, and our world.

The union in my opinion is unlikely to prevail. The typical Silicon Valley type of company is a caste system or a Great Chain of Being. The idea is that “leadership” is at the top of the hierarchy. Next in line are the smartest and most endorsed technical employees. Below the second tier are serf; that is, people who do work. The set up is efficient when viewed in terms of revenue, performance metrics, or other outputs deemed important to leadership. Thus, it is possible for some serfs to move up in the hierarchy. Certain “groups” of serfs have more difficulty moving up than others. Lawyers not skilled in technology face headwinds and may work from a trailer in a parking lot about a mile from Carpetland on Shoreline Drive. Others like a 17 year old whiz can go aloft carried by a hot air balloon of a technical breakthrough that produced a “big win” in the eyes of leadership.

The unionizers face a reality: Google wants compliant serfs. Dr. Timnit Gebru is the poster child for how Google responds to nails that stick up. The “no tech for apartheid” crowd found themselves free to operate a taco lunch wagon in Redwood City. And there are other examples.

The “We Are a Fighting Union” information says it wants:

Job protection
Fair pay for serfs
Flexible work conditions
Privacy protection
Racial and economic justice.

Okay, this is a great list for a discussion group in a high school sociology class. The five items are simply not part of the modern Silicon Valley-type company operating modality. Here’s what’s expected of serfs and what the Boards of Directors want to happen:

Screw up or are deemed not essential, a person will be terminated. The Meta email firings provide a good example of the modern management methods for personnel procedures
Pay is based on the tier and an “employee’s big wins.” No wins? You are out.
Work must be done within the “time” defined by leadership. This means a standard work day plus whatever additional hours are required and continuously over a seven day work week. Work, not the serf’s needs, comes first. Fail and the serf is pushed out.
Privacy does not exist. The idea is 24×7 surveillance with analytics identifying the non compliant.
Justice. Give me a break. Leadership defines the rules. Keep up and follow them or you will be in a taco truck in San Carlos.

Is this a good way to run a company? It depends on one’s point of view. From the leadership level, the answer is, “Absolutely.” From the mid-level, on the way up perspective, it’s not great, but it’s going to make me rich and I get nifty opportunities. From the serf level, it sucks. But throughout history, serfs have not been happy. Suck it up or quit strike me as the options.

Trust me. The creation of a union in a Silicon Valley type company is unlikely to result in the outcome the organizers and members expect.

Remember. Money and power in Silicon Valley type companies foster the emergence of Dark Age management methods.

Stephen E Arnold, June 5, 2026

Is the Oura Ring Thing a Handy Dandy Data Sucker Upper?

2026-05-29T17:50:10Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

I read a zero-slop article titled “Oura Says It Gets Government Demands for User Data. Will It Share How Many?” The Oura is a fitness device that gathers data about its wearer. The idea is that wearing a smart ring is better than wearing a big, clunky watch, toting a super sized mobile phone, or just being a person of interest with a black SUV following along as you go from home to Starbuck’s to your workplace in Managua.

The article includes some interesting factoids; for example:

In 2026, “… health wearable maker Oura became embroiled in a social media shitstorm after inking a deal with the Department of Defense and Palantir.”
…”many (if not most) companies design their systems to allow their staff to access user data, perhaps for troubleshooting customer issues or because it was the easiest and cheapest setup for a once cash-strapped startup.”
… “Oura data is not end-to-end encrypted.”
“Oura has sold over 5.5 million rings to date” …

But the most intriguing comment in the article, in my opinion, is this passage:

A spokesperson told me at the time [September 2025] that while Oura does not publish a transparency report, the company said it was “actively evaluating how to share aggregate data in a way that maintains security and does not introduce risk to our members.” It’s been eight months, dear reader. I recently reached out to Oura again to see if it would release a transparency report, and after several follow-up emails, the once-responsive Oura has not yet replied to any of my inquiries, or committed to releasing the numbers.

The pattern of taking an inquiry and offering a vague comment and then going radio silence is not unusual when [a] a firm is working with certain government entities, [b] lacks the PR savvy to spin a response that does not create more problems for a company, or [c] has something it does not want to be known by its 5.5 million customers, investigative journalists, or its staff.

Which is it? I have no clue. But like South Korea, each step in data acquisition seems to lead to an us versus them. The “us” are the organizations and professionals working for or in government entities. The “them” is any individual or group that allows paranoia or a threat to a mission to exist.

The longest journey begins with a single data broker. Where does it end? Ponder that.

Stephen E Arnold, June 5, 2026

One AI Bottleneck Workaround: From the WEF No Less

2026-05-31T17:46:08Z

The World Economic Forum discusses the barriers AI is finding when it comes to development: “AI Is Hitting A Wall. Here’s How We Rethink The Hardware To Break It.” When AI models grow larger that more time and energy needs to be spent moving data between memory and processors. It’s called “memory wall” and language-processing model grew 5,000 fold in size over four years.

Memory wall is a problem because large scale AI systems are increasing. This drives up costs and infrastructures. A second reason is that many valuable AI uses rely on fast decisions made locally instead of the cloud. Medical devices, autonomous vehicles, rescue drones, and more can’t rely on sharing information with data centers and waiting on responses. They need hardware that makes the AI more practical.

There are ways to overcome the AI memory wall bottleneck:

“…there are three main ways to ease this bottleneck: move computation closer to the data, draw on the brain’s event-driven information-processing method and use lower-precision or stochastic computing where exact arithmetic is unnecessary. Together, these approaches could support a new generation of AI hardware that is faster, more efficient and better suited to large-scale infrastructure and edge applications.”

These three options are the most powerful when they’re part of a single design solution. AI hardware that is designed in the future can’t rely on a single chip then fitting the algorithms on it as an afterthought. They need to be considered as part of the architecture during the design phase. Here’s what should be done:

“That is why hardware-algorithm co-design is becoming so important. Some workloads may benefit most from compute-in-memory; others may benefit from spiking networks and event-based sensing; and still others may rely on mixed-precision or stochastic methods. In many cases, the best solution may combine these approaches on the same platform. The larger implication is that the future of AI depends as much on hardware design as on model design. More efficient AI hardware could help contain the growing resource demands of large-scale systems while improving the safety and reliability of devices in the field.”

Interesting. But what if there is something other than the Google Transformer-centric method? The WEF will pivot, of course.

Whitney Grace, June 5, 2026

Microsoft May View AI As a Super App

2026-05-31T12:46:52Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

One of the simplest tricks in debate has a bunch of names. Game players like to say “pop up a level” or just “level up, dude.” Those with exposure to the rigors of traditional indexing say, “Just put it in a different cluster.” Silicon Valley types infused with the zeitgeist of incomprehensibility offer, “Do the meta play.” A rose by any other name still smells like an old age home on Sunday morning.

A great thinker wonders if a rose is actually a cat. Perhaps he can sell this concept and make lots of money? Thanks, MidJourney. Good enough.

The concept is that one has a thing and one wants to make that thing appear more profound, new, zippier, or imbued with qualities a person unfamiliar with “level upping” does not understand, see, or comprehend.

I want to point to an example described in “Exclusive: Microsoft Is Building a Super App That Combines Coding, Chat, and Other Copilot AI Tools.” (I like the exclusive part because by my count there are about a dozen outfits building super apps; however, as a dinobaby, I am generous. I think the “exclusive” is that no other big time real business news outfit has ever reported on Microsoft doing the clustering play or the “pop up a level, dude” tactic.)

Be that as it may, the “exclusive” write up reports:

The software giant is working on a one-stop shop that would connect its GitHub Copilot coding assistant, Copilot chat function, Copilot Cowork tool, and a new agentic workflow capability internally named Autopilot into a single app, according to two sources familiar with the project, who spoke on the condition of anonymity to discuss a platform that hasn’t yet been released.

Yep, a meta-app that will deliver a “cohesive product.” Isn’t this an admission that after three years of AI innovation, users of Microsoft software, hardware, and services are generally confused about what Copilot does what and how. The effort to data has caused a number of Microsoft users to want a way to turn off AI everything in Microsoft “experiences.” Some third parties have been paid big bucks to make the Copilot thing more obedient. But for most users, AI has begun to morph from “gee, this is cool” to “gee, this is a bit of a problem.”

The write up continues:

There may also be a toggle function for a user to go back and forth between their personal and enterprise 365 Copilots. A user will still be able to access their Copilots outside of the super app.

If this statement is indeed true, doesn’t this suggest that Microsoft is just foolin’? There will be many Copilots, not a single super app like the ones available from a Chinese outfit or from the alleged criminal Pavel Durov of Telegram and VKontakte fame. I won’t mention the others because then I will be revealing some of my upcoming lecture for the cyber fraud folks in Virginia in a few weeks. Some of my information, in my dinobabyish opinion, is indeed “exclusive.”

How lost in AI weirdness is Microsoft? May I suggest that this statement from the Fortune article helps answer the question? Here you go:

Microsoft in the past year has undergone one of the largest corporate reshuffles in its history that has included a string of high-profile departures and reorgs throughout its businesses. In April, it announced its first-ever employee buyout offer, aimed at its most long-tenured employees. At next week’s Build conference, Microsoft AI Chief Executive Mustafa Suleyman is expected to unveil new proprietary AI models. Suleyman, who once led consumer Copilot, has focused on models since the restructuring in March.

This is not recalibration in my view. This is scrambling and then trying to figure out how to put the eggs back in their shells.

Can Microsoft level up? Will the company survive the other super app developers? Will there be a reframing of the Microsoft AI strategy that includes security, defense against open source options, and the predatory instincts of Googzilla-type organizations in the US and from — yes, it is really true — other countries?

Confident in your answer? Hit those prediction markets.

Stephen E Arnold, June 4, 2026

eTools: Privacy and a New Swiss Search Service

2026-05-29T17:49:33Z

Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.

In one of my feeds, I spotted a link to a service more than 15 years old as I recall. I think the operator is Comcepta. The firm’s Web site explains enterprise metasearch.

There is a mobile, consumer facing version of the system. eTools.ch pops up an interesting message:

I know this is difficult to read. Here’s the text:

On this website we use cookies and similar functions to process end device information and personal data (e.g. such as IP-addresses or browser information). The processing is used for purposes such as to integrate content, external services and elements from third parties, statistical analysis/measurement, personalized advertising and the integration of social media. Depending on the function, data is passed on to up to 304 third parties and processed by them. This consent is voluntary, not required for the use of our website and can be revoked at any time using the icon on the bottom left.

I captured the names of the third party outfits this Swiss outfit works with. A person using this system on a mobile device will have some information shared with lots of companies. Here’s a snip from the master list of 304. I pulled the vendors from “G” to “L.” How many do you recognize? Also, in the screenshot of the cookie control panel, please, notice there is no option to disable “all” of these third party services.

GeoEdge
GfK GmbH
Glomex GmbH
Google Ads
Google Advertising Products
Google Analytics
Google General
Google Recaptcha
Grabit Interactive Media Inc. dba KERV Interactive
GumGum Australia, Inc.
GumGum, Inc.
Happydemics
Hearts and Science München GmbH
Human
Hybrid Adtech GmbH
ID5 Technology Ltd.
Impactify SARL
Improve Digital
Index Exchange Inc.
INFOnline GmbH
Innovid LLC
InMobi Technology Services Pte. Ltd.
Inskin Media LTD
Intercept Interactive Inc. dba Undertone
Invibes Group
IPONWEB GmbH
jsdelivr.com
Kairos Fire
Knorex
Krush Media LLC
Kupona GmbH
Kwanko
LIFT DSP LIMITED
LinkedIn
LinkedIn Ireland Unlimited Company
LiveIntent Inc.
LiveRamp
LoopMe Limited

I did not run any queries on the mobile service because I did not have one of my special devices with me. I am not sure how much storage these thrid party services consume. I am not sure I want to be “tracked.”

My recollection is that this eTools.ch service sends a user’s queries to open source indexes and some commercial services. Different deals are offered to gain access to a Web index outfit’s content. The results are retrieved and displayed to the user. The method is called “metasearch.” A similar service (originally developed by a finance type in New York City) is StartPage.com. (I am not sure who owns this service now.) Comcepta does provide some information about how its system works. I looked at the block diagram. The approach uses the same XML configuration file approach that Vivisimo did when it was in the search business.

The eTools.ch service emphasizes that it is “transparent.” I want to point out that the system has been around for more than a decade. What’s interesting is that Switzerland is one of the countries that harbors a number of private and privacy centric services. Banks come to mind as well as Proton’s services.

I want to point out that “privacy” in Switzerland does not pay much attention to what third party vendors do with information. The idea is that the vendor allegedly does not performl server-side profiling of users and doesn’t monetize query logs. What these third-party vendors do is not a problem for the regulators or the vendor.

The reason I posted this is to answer the question, “Where can I get a current list of third party vendors who might enter into a deal to help a Web site or service operator to generate some revenue?”

Here’s your answer: eTool.ch.

PS. I did not see “artificial intelligence” or “smart software” mentioned when I checked out the Web site.

Stephen E Arnold, June 4, 2026

Good Enough AI May Not Be Good Enough

2026-05-31T17:47:01Z

Rest Of World explains something that we all knew: “The Agentic Divide: Why “Good Enough” AI Isn’t Enough To Survive The New Economy.” What this means is that many AI chatbots are being released and some are better than others. That is to be expected, but no one expected there to be chatbot inequality and its economic consequences.

Chatbots are expected to take on monotonous tasks thus freeing up employees for other work. A problem, however, is that most chatbots still fail at basic facts and Big Tech companies are still launching agents to handle more complex tasks. Here is what will happen:

“As AI agents become more integrated into the economy, companies and entities that deploy them will benefit disproportionately compared to those that cannot, Nick Srnicek, a senior lecturer in digital economy at King’s College London, told Rest of World. ‘We will see new inequalities of access, scale, quality and trust: divides between those who have agents and those who don’t; those who have good agents and those who have bad agents; those who have many agents and those who have few agents; and those who can trust their agents and those who cannot,’ he said.”

China and Singapore apparently introduced frameworks to regulate AI to focus on accountability and safety. Chinese local governments are boasting single-employee companies that rely on AI chatbots. India is even embracing AI to cut costs and scale quicker.

Better chatbots are already advancing companies, saving them money, and advancing in the markets. Another problem is that if governments and companies rely on chatbots, they can be canceled at anytime without notice. There are also security concerns:

Matthew Sharp, a research affiliate at the Oxford Martin AI Governance Initiative [said]:

‘The same infrastructure can become a surveillance layer if the data flows, defaults, and oversight are wrong,” Sharp said. “The safeguards around consent, purpose limitation, auditability, and political independence would need to be real, not merely architectural.’”

My fear is that “good enough” is the new standard of excellence. A race to dumbing down. Outstanding.

Whitney Grace, June 4, 2026

Telegram AI: A Coocoon Sort of Emerging

2026-06-03T13:45:53Z

Telegram has continued to nudge its AI initiative forward. Unlike the infrastructure approach taken by US big AI tech outfits (BAITs in my lingo), Telegram has designs on a cheaper approach. The company wants to rent AI compute from data centers with unused cycles. Telegram Notes takes a look at the approach in a new article titled “The Telegram’s Cocoon AI: Not Yet Mature.” (“Mature” is a more sophisticated way of suggesting that the AI initiative is not yet aloft.) The company continues to make strategic and tactical shifts. Some of these are more dramatic that renting cycles; for example, dumping the TONcoin “name” for the original Telegram crypto called “GRAM.” Telegram is definitely an interesting organization.

Stephen E Arnold, June 3, 2026