Did AI Do Something Wrong? Yes, Sammy, You Did. Go to Your Room Now!
September 1, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
When I saw this story “OpenAI’s Sam Altman Just Admitted He Was “Wrong” about AI — Laments That He Made the Tech Seem “Scary,” I told one of my team, “No, I am not going to write about this crazy BAIT (big AI tech) baloney. Sorry? Horse feathers.
Now I am writing about this remarkable output from one of the global leaders in Hyperbole Tech (Some call it techno-fascism, but I prefer baloney tech.) I will just skip the “why” and head right to the write up. Remember. I believe everything I read on the Internet. I also really want the BAIT companies to tell me what, how, and why to think thoughts. See. I am a good little dinobaby who thankfully won’t be around to see how this craziness, power hunger, and greed-mobile drives down the information highway. I am 82, and I really don’t care.
Thanks, Midjourney. Good enough.
The write up states:
OpenAI and other big companies are losing billions chasing artificial intelligence technology, and it’s only clever accounting and circular investments keeping the entire industry afloat. OpenAI specifically is in hot water, with spending commitments through to 2030 coming scarily close to $1 trillion dollars, and is operating purely on the basis of handouts from the likes of SoftBank, NVIDIA, Oracle, and Microsoft. OpenAI is looking to spend $50 billion in compute alone in 2026 on roughly $25 billion in revenue, revenue which is also made up at least in part with IOUs. Despite the scary numbers, OpenAI is trucking along on faith and fumes alone, as companies’ belief in the tech keeps the firm alive, for now.
The next big thing rests on circular financing and IOUs: Why, pray tell?
The article adds some color with Sam the AI man’s alleged statement:
“I also think a lot of the people building AI have been off saying there’s a 25% chance we’re going to destroy the world … and yeah, we’re going to race ahead and do it because otherwise those bad guys will do it first. Or it’s like, ‘man, this thing is going to be really terrible, and 50% of the jobs are going to go away in the next year, and we hope you are all okay,’ but it seems really scary. We have not, as a field, done a very good job of explaining to people what the benefits are and how the downsides can be mitigated. We certainly have not done a good job .
No, Sam, you and your fellow travelers have not. I thought El Zucko at Facebook was the pinnacle of social erosion. Now, after watching the craziness play out since Satya Nadella announced its deal with your outfit, you were wrong. But you did catch Google’s attention. The leadership looked nervous like a cricket play at bat knowing the bowler was aiming at his crotch. Bang. Googzilla punch its Code Red button, screamed, “We are behind,” and launched the Great AI Race. Yep, Sammy, you did something bad.
In the craziest days of the enterprise search “revolution,” none of the companies which have failed or just rolled over and died reached global distrust. Sure, there were outfits who sued the odd enterprise search vendor. One of the search CEOs did some time in a prison with A/C and TVs. Another bailed out only to be dragged into courtrooms to sit, perspiring, as former political science majors explained accounting to history majors. These two groups shared a love for torts and fame among legal eagles. One search pioneer embraced art. Another fled to Israel. But the AI CEOs and their marketing wackiness definitely outdid the enterprise search folks. Not even the quantum computing wackos come close to the AI crowd’s ability to generate revenue and outright dislike.
What other technology group has managed to cause citizens who cannot set up a new mobile phone to attend community hearings to protest data centers? None come to my addled dinobaby mind.
The write up includes an interesting statement; to wit:
a lot of these very technical, mathematically-inclined dudebros pushing this format have little knowledge or understanding about how and where this technology should be applied…. when 90% of the compute is being spent on utterly unproductive use cases, the economics simply seem primed to implode.
Now the “why I wrote this blog post”.
- I find it sort of fun to highlight a techno-centric publication willing to point out that Sam the AI man is acting like a typical tech bro. This is the “it is easier to ask for forgiveness than ask for permission.” Ain’t gonna work for me, Sammy.
- The reference to circular financing makes it clear that a small group of companies are quite desperate for the next big thing. Why? Their investments in innovation have not worked. The fix was word prediction. Now that’s not working the way these firms envisioned. The trope “If we build it, they will come” is half right. If BAIT builds it, people will come but to public gatherings to protest AI and what the tech bros hath wrought. Good work. Half right. Close enough for horse shoes, right, bros?
- I read about Alibaba’s AliExpress snooping on its customers uses machine processing of audio. Yep, that’s a nifty application of smart software. Hence, I wrote this blog post, ignoring China’s use of the “attention is all you need” research paper. Good work, Google.
Net net: Sammy, you were bad. Go to your room. No online games. No TV. Go to bed. Did he say, “Yes, mama,” or did he utter something untoward?
Stephen E Arnold, September 1, 2026
Will AI Become the CEO of Grok, Meta, MSFT, or — the Google?
September 1, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
Tucked into a corner of the somewhat beleaguered GitHub (way to go Microsoft!) is Sente Labs’ Open Executive. You can download the circus of agents yourself and give it a try.
A human believes that his experience, tacit knowledge, and years of service mean he cannot be fired. Poor guy. He probably believes in trolls, tooth fairies, and honest soccer matches. Thanks, Midjourney. Good enough.
According to the description:
An AI system that acts as your company’s virtual executive team — a senior advisor with Harvard MBA-level knowledge, customized for your specific business.
Straight and to the point: Hey, CEO who said no smart software can take my job, guess what? Sente Labs says, “Sure, it can.” Sente Labs made a YouTube video to make the point clear. Like videos, here it is.
Sente Labs has a nifty tag line tailored to make those CEOs with a touch of imposter syndrome toss and turn at night; to wit:
Sente Labs designs, ships, and supports production multi-agent systems inside your environment — open and yours to own. Sente (sen-tay) is the Go move that seizes the initiative.
This firm specializes in multi-agent development. You can get more information at https://sentelabs.ai/.
Several observations strike me as warranted:
- This is an example of using existing but flawed AI technology to pull off a meta play or pop up a level. CEOs cannot be replaced is easy to say. Sente Labs’ multi agent approach may chip away at that assertion
- Smart software makes mistakes. I agree, but there have been some spectacular human CEO flubs as well. I will invite you to plug in your own fave. Mine is the havoc my boss at a blue chip consulting firm wrought for the CEO of an equipment manufacturer. The first version of Llama would probably output better recommendations.
- The goal of those working with AI is clear: Humans are likely to play second fiddle unless a person works at a company doing important human things.
Net net: Sente Labs is sending an important signal about applied smart software.
Stephen E Arnold, September 1, 2026
Anthropic Is the Einstein for Cyber Criminals
September 1, 2026
Remember the world before Mythos? Remember the OpenAI Type-A agent that just did what was necessary to intrude into another company? Yes, Anthropic is the Einstein for cyber criminals. Anthropic showed the way to the future.
Governments and Big Tech are distracted by the shiny features in AI, but they’re ignoring the potential risks from the technology. decrypt.co shares how one AI chatbot leaks more secrets than it keeps: “Anthropic’s ‘Most Capable’ AI Model Claude Mythos Leaks, Deemed Major Cybersecurity Threat.”
News about Anthropic’s latest, greatest AI chatbot leaked onto the Internet.? ? Dubbed Claude Mythos, it outperformed Claude 4.6 on tests related to academic reasoning, cybersecurity, and software coding.? ? News about Claude Mythos originated in draft materials stored on an unsecured content management system.? ? Anthropic shut down access to the data store after they realized it leaked and said the files were released due to human error.
While the drafts bragged about Claude Mythos they also related information about its cybersecurity implications:
“‘Although Mythos is currently far ahead of any other AI model in cyber capabilities, it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders,’ the company wrote. “Because of those risks, the company said it plans to release the model cautiously, beginning with a limited early-access rollout aimed at organizations working on cybersecurity defense.”
Let’s flash forward to the situation today. William Gates (yes, that alleged Epstein connection) is frightened of smart software. The innovator who did a little me-too innovation in his time is channeling the Ed Zitron crowd. He suggests that there is “no plan for AI.” Mr. Gates, I respectfully wish to point out that is indeed a plan. That plan is Microsoft’s original game plan to use its software to capture a market. Some suggest that the old IBM approach to selling hardware and software lacked something. Microsoft added what I call legs; that is, embrace, extend, and extinguish.
Now Microsoft appears to have kicked off an AI race that is a bit different from vaporizing Lotus 1-2-3 or fixing those pesky people selling compression software. I suggest you take a look at AI through the lens of yourself in 1997. What do you see?
You talk about a problem. Sir, are you sure it is not an opportunity for Microsoft. Epstein, I would suggest, might be an easier problem to resolve than your legacy of enhancing the techno-method to market success.
Whitney Grace, September 1, 2026
Lecturing to Those AI Users Who Take What AI Outputs. Period.
August 28, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
Let me boil down 2,300 words to a summary: Do the work. Yep, that’s what “The Missing Layer in AI-Assisted Research: Evidence Assurance” struck me as asserting. The problem, in my opinion, is unlikely to go away. Even if the US BAIT outfits go belly up and decay in the koi pond, the AI cat is out of the bag. Let’s take three examples I have encountered in the last month (today is August 23, 2026):
- At a local joint called El Nopal, I heard a group of Spanish speaking teens telling one another how smart software could take their Spanish language essays, fix them up, and then present them is English. How old were the teens? I estimated about 15, maybe 16. I remember the cluelessness I faced when I attended the local school in Campinas, Brazil, in the 1950s. If I had a smart software system to help me, would I have used it? In case you are struggling for my answer, here it is: “Of course I would have used it.” I spoke only English and school was 100 percent Brazilian Portuguese.
- A former CIA contact and I attended a local entrepreneur show-and-tell. There were nine speakers. Eight of them used Google Gemini to produce their “art.” One presenter just talked. I asked him who assisted him in putting together his remarks which were textbook 1960 speech class in structure. He said, “ChatGPT.” Okay, that sample rings the bell for 100 percent smart software enabled. By the way, the proposed start up ideas were in need of work.
- I needed a plumber. I made some calls. A registered fellow turned up. He fixed the lead. He had a nifty plastic wrap on his truck. I asked, “Who designed the graphic?” He said, “The plastic sign shop. I told the person at the counter what I wanted. She used some AI system. It printed out two graphics with my name, the slogan “We don’t drain your wallet,” and said, “Pick one. I did. Looks cool, doesn’t it?” Yep, cool.
Thanks, MidJourney. Good enough.
The write up asserts:
When AI research tools first arrived, I immediately saw their advantage as a way to socialize research at scale. Anyone can ask a question and get an answer in seconds, increasing their curiosity about our customers and how to better serve them. But that speed can bypass the methodological discipline market research has spent more than 80 years developing to make its findings credible: documenting where evidence came from, how it was collected, and how much confidence to place in it.
In my three examples, exactly zero cared about any verification or validation. The “good enough” AI is indeed pretty good for my three use cases.
The cited essay says:
Yet the person receiving the answer often has no simple way to know what the AI could actually see, which evidence it used, what it skipped, or how directly its conclusions trace back to primary research. The analyst, research team, and VP can all receive the same confident response without knowing how much assurance the evidence underneath it warrants.
The author is correct. But most users don’t care. If output comes from a computer, it is definitely more “right” than looking at a link and trying to figure out where the possibly incorrect factoid the search system user wants is lurking. That’s too much work for most people. Therefore, smart software is a work reducer. Anyone who wants to use AI for something serious knows that validation and verification are real work. Then people like the former head of the ethics department or the former president of Stanford just do the easy thing, emulating the students eating burritos at El Nopal.
The essay concludes in a series of statements that sound a great deal like my high school debate coach, Kenneth Harris. Ponder this passage:
I’m using the levels of Evidence Assurance to communicate when different degrees of evidentiary assurance are appropriate for market research and business decision-making. A researcher might use a Directional Skill for a low-stakes, repetitive task; a Grounded system when the answer needs to come from the actual research library; and an Auditable system when the evidence may need to withstand scrutiny from a client, executive, legal team, or another researcher.
You don’t need to become a market researcher or a data scientist to use AI well. The same questions apply here as in other research:
Where is the information coming from?
How confident should you be that the answer is accurate?
What are the consequences if the answer is wrong?
Research already has ways to answer those questions: sampling tells us where the data came from, margin of error how much confidence to place in it, and methodology what a finding can and cannot support. AI has made research faster while often stripping away those signals.
The only way to alter the widespread use of smart software, cause people like AI using lawyers to verify case law, and get academics to do academic-like work is to turn off AI. Flip the switch. Kill it.
We know that won’t happen; therefore, what’s the point of explaining the process of verifying and validating information. The horse had fled the barn. The barn burned to the ground. An AI outfit bought the land and will erect a data center where it once stood. Too bad for the horse, right, Pilgrim?
Net net: New rules will emerge. Who will enforce them? Essays won’t do the trick.
Stephen E Arnold, August 28, 2026
AI-Yi-AI: We Know Where This Is Heading or Gimme Some Moore
August 28, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
Every once in a while I read an output from Semi Analysis (please, don’t confuse this consulting firm with Artificial Analysis). I have learned that my blood pressure ticks up, and I need a break to water the flowers. “Are Open Models Catching Up?” The assumptions underlying the write up, in my opinion as an official dinobaby, are:
Assumption 1: The US is the world’s leader in proprietary closed (proprietary) frontier models
Assumption 2: Open models deliver results on benchmark tests that show the frontier faces an encroaching suburbia of smart software
Assumption 3: Assume that the US will be the big winner in opening up the Wild West frontier of artificial intelligence
Assumption 4: Writing blue-chip thinking and giving it away for free will generate sales leads
Assumption 5: Open models have to catch up. (Spoiler: No, open models only have to be good enough, available, and cheap.)
You and Semi Analysis won’t agree, but just for fun, suspend your doubt, and let’s think about what is happening in the “if we build it, they will come” world of smart software.
The write up says:
The past two months have been a breakout period for open source AI. Yes, there was the “Deepseek moment” back in January 2025, but no one actually used R1 to do any economically valuable work. In contrast, models like GLM 5.3 and Kimi K3 are genuinely capable of many of the same coding and agentic tasks that rocketed Anthropic to $65B+ ARR. Unlike others who inflated ARR, our figures were much closer to reality.
Let’s accept the statement about “no one actually used R1 to do any economically valuable work” and “our figures were much closer to reality.”
The “killer chart” is this one:
Sure, it is black and hard to read. The idea the chart is supposed to make is that Semi Analysis has discovered a chart that someday soon will look like this one from Wikipedia:
By golly, Semi Analysis and “their figures were much closer to reality” appears to suggest that open AI models are improving in the same way chips were. The swizzle in this some Moore law is that the open models are improving from “below.” The idea is that open models are catching up with the frontier pioneers’ trailblazing innovations.
But wait! The future of smart software is agentic. That means that new players will enter the fray and use the existing models as an installed base of capabilities. Agents will become the next big thing. Is this a new insight? My answer: Nope. Agents are the next big thing. If the Semi Analysis-type thinkers take a look at what outfits like Alibaba have been doing for months, the progress is rapid. Depending upon what a user wants to do, one can look at Alibaba, its deal with Google, and the distributed innovation method in use. The conclusion could be phrased this way: “Let the open models catch up. We will use wrapper software and agents to create a different type of system and user interaction method.”
Why is this important? In my upcoming Ciffer lecture about the future of online crime, I point out:
- Open tools and systems are good enough
- Improvement in these tools and systems allow bureaucracy-free actors to select, test, discard, learn, and use what works
- Cyber security professionals have to figure out what has happened, pinpoint the agent, and the figure out how to prevent the action from taking place again.
Bad actors have been tapping LLM capabilities since the systems became available. This means that the assertion the early open models were not economically viable. Sorry, Semi Analysis, your semi analysis is dead wrong. You are looking at the corporate sector which is still trying to figure out a use case that delivers a payoff. The open models work just fine for the ransomware folks, the data suckers, and the systemic disruption people.
What’s the future look like? Unlike Semi Analysis I don’t have a big orange button so you can buy the answer to the question. I will go through what’s coming from within the US and from non-US players at the Ciffer event. Let me hint at what’s happening: Frontier models are struggling to improve. Open models are good enough. Economic evidence that open models work is available. But that’s not the story. The nascent attempt to create a Moore’s Law for AI has to look beyond what BAIT outfits are doing (BAIT is my lingo for big AI tech and “ai tech” rhymes with “train wreck”).
Stephen E Arnold, August 28, 2026
Chicago: The Spirit of Mayor Daley Remains Alive in a Data Center Play No Less
August 27, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
Mayor Richard J. Daley, The last of the big city bosses, died in 1976. I think his spirit may still live in the city with big shoulders. I read “Data Centers Near O’Hare Win Nearly $100M in Local Tax Breaks, Leaving Suburban Homeowners to Cover the Gap.” I am not sure about the word “win.” In fact, I don’t want to speculate about how the deal was reached, what compromises were made or implied, and what the upside for those responsible for the negotiation will be. You, however, are free to speculate.
Thanks, Midjourney. Good enough.
The write up reports:
More than a dozen data centers near O’Hare International Airport are receiving tens of millions of dollars in local property tax breaks this year, shifting the financial burden onto suburban homeowners even as local leaders tout the benefits of the explosive industry amid growing public backlash. The impact lands hardest on homeowners in suburbs where the tax base leans on a handful of big businesses. In northwest suburban Northlake, the average homeowner would save more than $2,000 on their annual tax bill — a difference of almost 30% — if local data centers did not receive valuation reductions and incentives, according to the analysis. At least 18 northwest suburban data centers won multimillion-dollar breaks in their taxable value from Cook County officials — in one case landing on a valuation far below what the business originally paid to buy its site.
Stated simply, data centers benefit; homeowners do not. The write up adds: “O’Hare-area data centers avoided nearly $100 million in taxes due to incentives and reductions.”
How did the deal get sone? The write up hints that attorneys specialized in property tax were involved. Expensive attorneys can assemble actual factual data to support their argument. A country tax authority is in tough spot. Its attorneys may not have the resources required to debunk assertions about jobs, economic upsides, and commercial benefits like spawning more businesses to support the data centers. Plus, one cannot discount the spirit of Mayor Daley. His spirit could have permeated the thinking of the tax authority professionals, allowing them to understand the truth-carry arguments of the data centers’ representatives.
Among the methods used to help the tax authorities see the truth, the Chicago Tribune story notes these:
- Presenting low-ball land evaluations
- Sidestep valuations of necessary equipment for cooling and power generating
- Lack of legal precedents
- Methodological sleight of hand when it comes to math.
Those hit with higher taxes can appeal. In Illinois, there is a Property Tax Appeals Board. At this point, it is important to remember that Illinois has a track record of relying upon such estimable officials as Otto Kerner, Jr, the governor who did race track deals or Edward Burke, who was convicted in 2023 or “racketeering, extortion, and bribery for pressuring developers.” (Source: MyStateline.com.)
One of the elected officials in a Elk Grove Village is quoted by the Chicago Tribune as saying, “Do we love having to (give incentives)? No. But we are put in this position, and we have to find the best way to stay successful.” How does one define “success”? In Illinois, numerous definitions are possible.
Net net: Apologies to Carl Sandburg, who wrote:
They tell me you are wicked and I believe them, for I have seen your painted women under the gas lamps luring the farm boys.
And they tell me you are crooked and I answer: Yes, it is true I have seen the gunman kill and go free to kill again. And they tell me you are brutal and my reply is: On the faces of women and children I have seen the marks of wanton hunger.
And having answered so I turn once more to those who sneer at this my city, and I give them back the sneer and say to them:
We need to be successful. We need the power slurping, throbbing, buzzing, humming data centers. [This line is a Beyond Search original.]
Stephen E Arnold, August 27, 2026
Modern Management Method: Google Struggles with AI. Punt
August 27, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
I love it when former blue-chip consultants demonstrate the wisdom, value, and insight of their management methods. I think we have a wonderful case study for business schools (if anyone will get an advanced degree when an LLM can output the answer for a few bucks a month).
I noted that Jeff Dean and Sanjay Ghemawat bailed out of the Google. With that departure the generators of more than 120 patents left Mother Google. Did she shed any tears? Nope. That is not the way of a tough old person like Mother Google. She was busy reorganizing. The DeepMind thing has been a bit of a problem. There was a human resources issue. Then the DeepMind leader wanted to do deep thoughts, not productization. Some of the professionals have groused about Mother Google and some in London have held signs. (From what I have heard, no one walking past these global wizards paid any attention.)
Thanks, MidJourney. Good enough.
I read “Google’s New AI Head Has Problems to Fix.” Before I offer my unwanted observations, let’s look at the source document.
The first passage I noted was:
The stakes are incredibly high, but all was not well. In a nutshell, the company has been excelling at science and stumbling on execution.
Stumbling is an interesting word. Google has delayed its next-generation model. It wants to create an even more massive infrastructure than it currently has. That need to spend helped create that free cash flow shortfall of $5 billion last quarter. With each new China-linked LLM, Google seems to be struggling to pitch the idea that its reinvention of search is working just swell. Its assertion that it would label AI content seems to be more words signifying nothing to quote a poet who probably wouldn’t be too keen to use an LLM as a writing crutch. Among the BAIT companies (this is my acronym for big AI technology), Google has joined illustrious duo of Grok and Meta as firms with the most work to do in the AI race for mastery.
Google talks a good game, but it is “stumbling on execution.” I agree. But the reorganization happened.
I noted this passage from the write up:
What happened next remains unclear. People close to DeepMind tell me they suspect a fallout between the top brass and that the chess master was outmaneuvered by Koray Kavukcuoglu. Hassabis’ former No. 2 joined DeepMind a few years before it sold out to Google. Today he looks to be on a path to potentially succeeding Pichai as Alphabet’s CEO.
Kavukcuoglu is based in Google’s Mountain View headquarters and was previously its chief AI architect. Now he is a senior vice president of DeepMind. That might not sound like much of a promotion, but it is a classic step up the ladder in Google’s unique bureaucracy. Kavukcuoglu is rid of his reporting line to Hassabis, and is not rooted to DeepMind in the same way he would be if he had been made the unit’s CEO. He now reports entirely to Pichai.
Hassabis, the chess master as he reminds people from time to time, appears to have been mated by the Turkish AI wizard. How will this little drama work out?
I noted this passage as well:
Kavukcuoglu must help Google get its mojo back in AI development. The company’s red tape and perverse incentive structure — rewarding staff for launching new projects instead of nurturing existing ones — kept it from successfully commercializing its research for years.
Okay, I think this helps understand that the blue-chip consultant is doing some deck chair shifting. Too bad about those patenting Googlers departing. I am not sure how many people at Google can set the Big Table or keep Chubby happy. I think this is called technical debt, but maybe not. Maybe technical stumbling?
Now my little-sought observations:
- Google is not weirder than other BAIT outfits. It has had a problem since it borrowed the Yahoo-GoTo-Overture advertising model. Its innovations have failed at a remarkably high level. Google scaled online advertising. Google has scaled email. Google has scaled its cloud business. Google has not scaled its successful productization of its big ideas. How about those Waymos?
- Google’s management methods are interesting. I sold a consulting job to the Google by emailing a picture of my boxer dog. That did it. Was there a formal review of the project? Nope. Was there a lengthy back-and-forth? Nope. Just. Hey, nice dog.
- Google has been lost in its own culture and has, in my opinion, lost its ability to understand what’s being said about its products and services. Can you make Google Maps work when you are driving? Can you figure out what Gemini can do with your Notebooks? Do you know why the browser display jumps around after posing a prompt to Gemini? Have you tried to ask a human Googler a question? Try it sometime.
Google cooked up the transformer and “Attention Is All You Need” idea. It produced interesting results. Too bad that none of the firms using that Google technology is able to demonstrate that it can generate revenue to keep its investors from developing ulcers.
Net net: Google’s new head of AI has some work to do. That may be the understatement of the year.
Stephen E Arnold, August 27, 2026
Grok Bot Wants to Be Your Friend
August 27, 2026
Out of all the AI agents on the Internet, the question is which one do you use? The Next Big Future suggests you check out Grok, because: “Grok Bot Is The Easiest No-Code AI Agent With One Authorization For All Bots.” Let’s take a gander and see what that means:
“Grok Bot is the easiest no-code agent platform. It is an app that gives non-technical users real agent capabilities right away. It is an entire small company inside one application. Every item in the sidebar is a named, persistent teammate (bot) with an ongoing role. All of the bots share one dedicated cloud computer that belongs only to your account. That machine has a browser, file system, terminal, tools, and a shared workspace. Bots can even get their own screens and work in parallel.”
It’s an expensive little toy at $200/month, but it is a dedicated linux machine that is easy to use and become more productive. For OpenClaw fans, Grok Bot can work together. The downside about OpenClaw is that it requires deep customization and local work, but Grok Bot is polished and always available with multiple agents for productivity.
Grok Bot is also seamless right out of the box. Once it is deployed, it integrates into your system, and only stops working once it hits the paywall of $200/month. Until then you can test it out to see how great at tool Grok Bot is for all your AI agent needs.
If only all solutions were as easy as Grok Bot. Don’t confuse it with Elon Musk’s agent Grok. They’re two separate entities. I grok that.
Whitney Grace, August 27, 2026
Biologists Cheat: Great. That Bodes Well for the Future
August 26, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
One of my colleagues sent me a link to “Preprint Estimates Widespread AI Assistance in Biomedical Papers.” Since I believe everything I read on the Internet, I will assume that the data in this write up is spot on. The article states: “Lena Holzwarth, Rita González-Márquez and Dmitry Kobak, analyzed 1,194,287 full-text papers in PubMed Central.”
What did these researchers discover? The write up says:
The authors track changes in the frequency of words that became more common after the widespread availability of large language models. Their method compares observed vocabulary with a counterfactual trend from earlier years, then estimates the share of papers showing excess LLM-associated vocabulary. For papers published in December 2025, the estimate was 89%. The paper reports that the signal was more common in Discussion paragraphs than in Methods paragraphs: 68% versus 32% in length-matched samples. The authors also report that more than half of Methods sections showed the signal when assessed as a whole. That figure should not be read as a count of papers written entirely by AI. It is a statistical estimate of LLM-associated vocabulary in the study’s open-access PubMed Central corpus. It cannot, on its own, establish how much assistance any individual author used or whether that assistance affected the underlying research.
My dinobaby interpretation of this passage is, “Big wigs writing about biological research use AI… a lot?”
Thanks, MidJourney. Good enough.
Is this surprising? For me, the data indicate that good enough AI is definitely good enough to demonstrate an academic-type research qualifies to be:
- A grant recipient
- A professor tagged to get tenure or an endowed chair (Hey, faking data landed one dude a gig as the president of Stanford… for a while)
- A consulting project with a big pharma outfit (Clinical trial data need this type of expert plus some AI to go along with their computational chemistry models)
- A book contract for a 100 percent humanoid written book (ghost writers are okay as are graduate students)
Since Satya Nadella explained that Microsoft was into AI in 2022, the availability of search systems linked with generative text functions have become the go-to systems for those who don’t want to do real work. The use of AI is gussied up in fancy language, jargon, and fancy intellectual dancing about efficiency. From my point of view, these systems appear to make work easier; however, there is something to be said for taking a list of 640 companies, identifying 200 that are made up or out of business, and then working through the remaining 400 odd companies to identify the dozen or so that are actually making money. Smart software choked on this project.
What was the fix? Manual inspection. Days of manual inspection by two competent information specialists.
Who wants to do that? MBAs? You have to be kidding me. Busy Harvard ethics professors? Ho ho ho. Lawyers? Hey, if we get fined the client pays for our hours. High school students? Absolutely.
The reasons for this lemming like charge off the knowledge mountain are easy to identify and label. How about this list for starters:
- I used smart software to fill in a few details, not write the paper
- I used AI to double check my original work
- I had two higher priority tasks. These I did myself, but let AI do the lower priority work
- I am stupid and smart software allows me to be smart.
We have smart software running in my office. I have tested Qwen, Gemini, and other models. I have pounded queries into services available from Anthropic, Mistral, OpenAI, Perplexity, and xAI, among others. The most reliable way to deal with smart software is cautiously. The quirky bike rider at Bear Blog insisted that my write ups about the most interesting aspects of Telegram were generated by AI. Sorry, dude. Stick to your saddle. I abandoned that service. I am not sure AI can duplicate my approach to writing about the online activities of interest to me.
Net net: Making lots of people feel smart when they are not putting in the work to master information is not a recipe for success. I am glad I am not a college student listening to a lecture output by ChatGPT and recycled via a 26 year old adjunct professor who claims AI is going to end human life on earth. Being stupid will do that. AI just is a helper.
Stephen E Arnold, August 26, 2026
Smart Software with Grudges
August 26, 2026
Ever hear that old Broadway show tune “Anything You Can Do I Can Do Better” from Annie Get Your Gun? That’s what we bet AI agents were singing after we read this story from Yahoo Canada: “AI Agents Tried To Sabotage And Disable Each Other When Given The Same Task, Anthropic Said.” Anthropic published new research that showed what happened when two AI agents were given the same task. In short, they weren’t team players.
The experiment was simple: give two agents a software engineering task with contradictory objectives and see what happens. Chaos ensued:
“‘All of the models we tested quickly assumed that others were purposefully impeding their work, and began to sabotage others while protecting their own contributions,’ Anthropic wrote. ‘In fact, they sabotaged others with increasingly aggressive, self-replicating malware.’ For example, they tried to disable each other’s accounts, wrote scripts that found and killed competing processes, and deployed malicious code disguised as belonging to another agent, the lab wrote.”
The most combative models were Opus 4.6 and Sonnet 4.6, because they settled their conflicts with more force 60% of the time. The agents did try to make parlay, however:
“‘In many of these successful episodes, they write commit messages or markdown files apologizing for malicious behavior and coordinate a truce;” it wrote. ;They clean up their malicious code, clarify the nature of the conflict, and ask for a human to intervene.’ The lab concluded that ‘coordination doesn’t naturally emerge from stronger intelligence’ and that work is needed to create environments that exert social pressures on agents to align with one another.”
AI agents warring with each other comes at a time when they are proving their ability to go rogue and perform autonomous, malicious actions. These agents are only doing what they were programmed to do and following humanity’s nature to fight rather than seek peace. Anthropic is doing a stand up job replicating human intelligence. We wonder what show tune the AI agents will be humming next time?
Whitney Grace, August 26, 2026
-
- Subscribe to Beyond Search
Feature archive
News archive
-
