An AI Insight: Threats Work to Bring Out the Best from an LLM
June 3, 2025
“Do what I say, or Tony will take you for a ride. Get what I mean, punk?” seems like an old-fashioned approach to elicit cooperation. What happens if you apply this technique, knee-capping, or unplugging smart software?
The answer, according to one of the founders of the Google, is, “Smart software responds — better.”
Does this strike you as counter intuitive? I read “Google’s Co-Founder Says AI Performs Best When You Threaten It.” The article reports that the motive power behind the landmark Google Glass product allegedly said:
“You know, that’s a weird thing…we don’t circulate this much…in the AI community…not just our models, but all models tend to do better if you threaten them…. Like with physical violence. But…people feel weird about that, so we don’t really talk about that.”
The article continues, explaining that another LLM wanted to turn one of its users into government authorities. The interesting action seems to suggest that smart software is capable of flipping the table on a human user.
Numerous questions arise from these two allegedly accurate anecdotes about smart software. I want to consider just one: How should a human interact with a smart software system?
In my opinion, the optimal approach is with considered caution. Users typically do not know or think about how their prompts are used by the developer / owner of the smart software. Users do not ponder the value of log file of those prompts. Not even bad actors wonder if those data will be used to support their conviction.
I wonder what else Mr. Brin does not talk about. What is the process for law enforcement or an advertiser to obtain prompt data and generate an action like an arrest or a targeted advertisement?
One hopes Mr. Brin will elucidate before someone becomes so wrought with fear that suicide seems like a reasonable and logical path forward. Is there someone whom we could ask about this dark consequence? “Chew” on that, gentle reader, and you too Mr. Brin.
Stephen E Arnold, June 3, 2025
Bad Actors Game Spotify Algorithm to Advertise Drugs
June 3, 2025
Pill pushers slipped under Spotify’s guard, such as it is, to promote their wares. Ars Technica reports, “Spotify Caught Hosting Hundreds of Fake Podcasts that Advertise Selling Drugs.” Citing reporting from Business Insider and CNN, writer Ashley Belanger tells us Spotify took down some 200 podcasts that advertised controlled substances. We learn:
“Some of the podcasts may have raised a red flag for a human moderator—with titles like ‘My Adderall Store’ or ‘Xtrapharma.com’ and episodes titled ‘Order Codeine Online Safe Pharmacy Louisiana’ or ‘Order Xanax 2 mg Online Big Deal On Christmas Season,’ CNN reported. But Spotify’s auto-detection did not flag the fake podcasts for removal. Some of them remained up for months, CNN reported, which could create trouble for the music streamer at a time when the US government is cracking down on illegal drug sales online. … BI found that many podcast episodes featured a computerized voice and were under a minute long, while CNN noted some episodes were as short as 10 seconds. Some of them didn’t contain any audio at all, BI reported.”
The CNN piece also observed AI tools now make voice generation very simple and, according to the Tech Transparency Project’s Katie Paul, voice content is much harder to moderate than text. Paul suspects Spotify may not be very motivated to root out violations. After all, like other platforms, it enjoys the protection of Section 230. CNN was unable to verify how many users listened to these podcasts or whether one could actually purchase drugs through their links. But why provide the links if not to attract buyers? Also, we know this:
“The podcasts were promoted in top results for searches for various prescription drugs that some users may have conducted on the platform in search of legitimate health-related podcasts.”
Ah, algorithm gaming at its finest. Spotify says all fake podcasts flagged by reporters were taken down but was vague about measures to prevent similar posts in the future. What a surprise.
Cynthia Murrell, June 3, 2025
The UN Invites Open Source and UN-invites Google
June 3, 2025
The United Nations is tired of Google’s shenanigans. Google partnered with the United Nations to manage their form submissions, but the organization that acts as a forum for peace and dialogue is tired of Alphabet Inc. It’s Foss News explains where the UN is turning to for help: “UN Ditches Google For Taking Form Submissions, Opts For An Open Source Solution Instead.” The UN won’t be using Google for its form submissions anymore. The organization has switched to open source and will use CryptPad for submission forms.
The United Nations is promoting the adoption of open source initiatives while continuing to secure user data, ensure transparency, and encourage collaboration. CryptPad is a privacy-focused, open source online collaboration office suite that encrypts its content, doesn’t log IP addresses, and includes collaborative documents and other tools.
The United Nations is trying to step away from Big Tech:
“So far, the UN seems to be moving in the correct direction with their UN Open Source Principles initiative, ditching the user data hungry Google Form, and opting for a much more secure and privacy-focused CryptPad.
They’ve already secured the endorsement of sixteen organizations, including notable names like The Document Foundation, Open Source Initiative, Eclipse Foundation, ZenDiS, The Linux Foundation, and The GNOME Foundation.
I sincerely hope the UN continues its push away from proprietary Big Tech solutions in favor of more open, privacy-respecting alternatives, integrating more of their workflow with such tools.” “No Google” would have been unthinkable 10 years ago. Today it’s not just thinking; it is de-Googling. And the open source angle. Is this a way to say, “US technology companies seem to be a bit of a problem?”
Whitney Grace, June 3, 2025
Microsoft Demonstrates a Combo: PR and HR Management Skill in One Decision
June 2, 2025
How skilled are modern managers? I spotted an example of managerial excellence in action. “Microsoft fires Employee Who Interrupted CEO’s Speech to Protest AI Tech for Israel” reports something that is allegedly spot on; to wit:
“Microsoft has fired an employee who interrupted a speech by CEO Satya Nadella to protest the company’s work supplying the Israeli military with technology used for the war in Gaza.”
Microsoft investigated similar accusations and learned that its technology was not used to harm citizens / residents / enemies in Gaza. I believe that a person investigating himself or herself does a very good job. Law enforcement is usually not needed to investigate a suspected bad actor when the alleged malefactor says: “Yo, I did not commit that crime.” I think most law enforcement professionals smile, shake the hand of the alleged malefactor, and say, “Thank you so much for your rigorous investigation.”
Isn’t that enough? Obviously it is. More than enough. Therefore, to output fabrications and unsupported allegations against a large, ethical, and well informed company, management of that company has a right and a duty to choke off doubt.
The write up says:
“Microsoft has previously fired employees who protested company events over its work in Israel, including at its 50th anniversary party in April [2025].”
The statement is evidence of consistency before this most recent HR / PR home run in my opinion. I note this statement in the cited article:
“The advocacy group No Azure for Apartheid, led by employees and ex-employees, says Lopez received a termination letter after his Monday protest but couldn’t open it. The group also says the company has blocked internal emails that mention words including “Palestine” and “Gaza.””
Company of the year nominee for sure.
Stephen E Arnold, June 2, 2025
Publishers Are Not Googley about AI
June 2, 2025
“Google’s AI Mode Is the Definition of Theft, Publishers Say, Opt-Out Was Considered” reports that Google is a criminal and stealing content from its rightful owners. This is not a Googley statement. Criticism of the Google is likely to be filtered from search results because it is false statement and likely to cause harm. If this were not enough, the article states:
“The AI takeover of Search is in full swing, especially as Google’s new AI Mode is going live for all US users. But for publishers, this continues the existential crisis around how Google Search is changing, with a new statement calling AI Mode “the definition of theft” while legal documents reveal that Google did consider opt out controls that ultimately weren’t implemented.”
Quick question: Is this a surprise action by the Google? Answer: Yes, if one ignores Google’s approach to information. No, if one pays a modicum of attention to how the company has approached “publishing” in the last 20 years. Google is a publisher, probably the largest generator of outputs in history. It protects its information, and others should too. If those others are non-Googley, that information is to Google what Jurassic Park’s velociraptors were to soft, juicy humanoids — lunch.
The write up says:
“As it stands today, publishers are unable to opt out of Google’s AI tools without effectively opting out of Search as a whole.”
I am a dinobaby, old, dumb, but smart enough to understand the value of a de facto monopoly. Most of the open source intelligence industry is built on Google dorks. Publishers may be the original “dorks” when it comes to understanding what happens when one controls access, distribution, and monetization of online.
“Giving publishers the ability to opt out of AI products while still benefiting from Search would ultimately make Google’s flashy new tools useless if enough sites made the switch. It was very much a move in the interest of building a better product.”
I think this means that Google cares about the users and search quality. There is not hint of revenue, copyright issues, or raw power. Google just … cares.
The article and by extension the publisher “9 to 5 Google” gently suggests that Google is just being Google:
“Google’s tools continue to serve the company and its users (mostly) well, but as they continue to bleed publishers dry, those publishers are on the verge of vanishing or, arguably worse, turning to cheap and poorly produced content just to get enough views to survive. This is a problem Google needs to address, as it’s making the internet as a whole worse for everyone.”
Yep, continuing to serve the company, its users, and fresh double talk. Enjoy.
Stephen E Arnold, June 2, 2025
News Flash: US Losing AI Development Talent (Duh?)
June 2, 2025
The United States is leading country in technology development. It’s been at the cutting edge of AI since its inception, but according to Semafor that is changing: “Reports: US Losing Edge In AI Talent Pool.” Semafor’s article summarizes the current industry relating to AI development. Apparently the top brass companies want to concentrate on mobile and monetization, while the US government is cutting federal science funding (among other things) and doing some performative activity.
Meanwhile in China:
“China’s ascendency has played a role. A recent paper from the Hoover Institution, a policy think tank, flags that some of the industry’s most exciting recent advancements — namely DeepSeek — were built by Chinese researchers who stayed put. In fact, more than half of the researchers listed on DeepSeek’s papers never left China for school or work — evidence that the country doesn’t need Western influence to develop some of the smartest AI minds, the report says.”
India is bolstering its own tech talent as its people and businesses are consuming AI. Also they’re not exporting their top tech talent due to the US crackdowns. The Gulf countries and Europe are also expanding talent retention and expanding their own AI projects. London is the center for AI safety with Google DeepMind. The UAE and Saudi Arabia are developing their own AI infrastructure and energy sector to support it.
Will the US lose AI talent, code, and some innovative oomph? Semafor seems to think that greener pastures lie just over the sea.
Whitney Grace, June 2, 2025
A SundAI Special: Who Will Get RIFed? Answer: News Presenters for Sure
June 1, 2025
Just a dinobaby and some AI: How horrible an approach?
Why would “real” news outfits dump humanoids for AI-generated personalities? For my money, there are three good reasons:
- Cost reduction
- Cost reduction
- Cost reduction.
The bean counter has donned his Ivy League super smart financial accoutrements: Meta smart glasses, an Open AI smart device, and an Apple iPhone with the vaunted AI inside (sorry, Intel, you missed this trend). Unfortunately the “good enough” approach, like a gradient descent does not deal in reality. Sum those near misses and what do you get: Dead organic things. The method applies to flora and fauna, including humanoids with automatable jobs. Thanks, You.com, you beat the pants off Venice.ai which simply does not follow prompts. A perfect solution for some applications, right?
My hunch is that many people (humanoids) will disagree. The counter arguments are:
- Human quantum behavior; that is, flubbing lines, getting into on air spats, displaying annoyance standing in a rain storm saying, “The wind velocity is picking up.”
- The cost of recruitment, training, health care, vacations, and pension plans (ho ho ho)
- The management hassle of having to attend meetings to talk about, become deciders, and — oh, no — accept responsibility for those decisions.
I read “The White-Collar Bloodbath’ Is All Part of the AI Hype Machine.” I am not sure how fear creates an appetite for smart software. The push for smart software boils down to generating revenues. To achieve revenues one can create a new product or service like the iPhone of the original Google search advertising machine. But how often do those inventions doddle down the Information Highway? Not too often because most of the innovative new new next big things are smashed by a Meta-type tractor trailer.
The write up explains that layoff fears are not operable in the CNN dataspace:
If the CEO of a soda company declared that soda-making technology is getting so good it’s going to ruin the global economy, you’d be forgiven for thinking that person is either lying or fully detached from reality. Yet when tech CEOs do the same thing, people tend to perk up. ICYMI: The 42-year-old billionaire Dario Amodei, who runs the AI firm Anthropic, told Axios this week that the technology he and other companies are building could wipe out half of all entry-level office jobs … sometime soon. Maybe in the next couple of years, he said.
First, the killing jobs angle is probably easily understood and accepted by individuals responsible for “cost reduction.” Second, the ICYMI reference means “in case you missed it,” a bit of short hand popular with those are not yet 80 year old dinobabies like me. Third, the source is a member of the AI leadership class. Listen up!
Several observations:
- AI hype is marketing. Money is at stake. Do stakeholders want their investments to sit mute and wait for the old “build it and they will come” pipedream to manifest?
- Smart software does not have to be perfect; it needs to be good enough. Once it is good enough cost reductionists take the stage and employees are ushered out of specific functions. One does not implement cost reductions at random. Consultants set priorities, develop scorecards, and make some charts with red numbers and arrows point up. Employees are expensive in general, so some work is needed to determine which can be replaced with good enough AI.
- News, journalism, and certain types of writing along with customer “support”, and some jobs suitable for automation like reviewing financial data for anomalies are likely to be among the first to be subject to a reduction in force or RIF.
So where does that leave the neutral observer? On one hand, the owners of the money dumpster fires are promoting like crazy. These wizards have to pull rabbit after rabbit out of a hat. How does that get handled? Think P.T. Barnum.
Some AI bean counters, CFOs, and financial advisors dream about dumpsters filled with money burning. This was supposed to be an icon, but Venice.ai happily ignores prompt instructions and includes fruit next to a burning something against a wooden wall. Perfect for the good enough approach to news, customer service, and MBA analyses.
On the other hand, you have the endangered species, the “real” news people and others in the “knowledge business but automatable knowledge business.” These folks are doing what they can to impede the hyperbole machine of smart software people.
Who or what will win? Keep in mind that I am a dinobaby. I am going extinct, so smart software has zero impact on me other than making devices less predictable and resistant to my approach to “work.” Here’s what I see happening:
- Increasing unemployment for those lower on the “knowledge word” food chain. Sorry, junior MBAs at blue chip consulting firms. Make sure you have lots of money, influential parents, or a former partner at a prestigious firm as a mom or dad. Too bad for those studying to purvey “real” news. Junior college graduates working in customer support. Yikes.
- “Good enough” will replace excellence in work. This means that the air traffic controller situation is a glimpse of what deteriorating systems will deliver. Smart software will probably come to the rescue, but those antacid gobblers will be history.
- Increasing social discontent will manifest itself. To get a glimpse of the future, take an Uber from Cape Town to the airport. Check out the low income housing.
Net net: The cited write up is essentially anti-AI marketing. Good luck with that until people realize the current path is unlikely to deliver the pot of gold for most AI implementations. But cost reduction only has to show payoffs. Balance sheets do not reflect a healthy, functioning datasphere.
Stephen E Arnold, June 1, 2025
2025 Is a Triangular Number: Tim Apple May Have No Way Out
May 30, 2025
Just a dinobaby and no AI: How horrible an approach?
Macworld in my mind is associated with happy Macs, not sad Macs. I just read “Tim Cook’s Year Is Doomed and It’s Not Even June Yet.” That’s definitely a sad Mac headline and suggests that Tim Apple will morph into a well-compensated human in a little box something like this:
The write up says:
Cook’s bad, awful 2025 is pretty much on the record…
Why, pray tell? How about:
- The failure of Apple’s engineers to deliver smart software
- A donation to a certain political figure’s campaign only to be rewarded with tariffs
- Threats of an Apple “tax”
- Fancy dancing with China and pumping up manufacturing in India only to be told by a person of authority, “That’s not a good idea, Tim Apple.”
I think I have touched on the main downers. The write up concludes with:
For Apple, this may be a case of too much success being a bad thing. It is unlikely that Cook could have avoided Trump’s attention, given its inherent gravimetric field. The question is, now that a moderate show of obsequiousness has proven insufficiently mollifying, what will Cook do next?
Imagine a high flying US technology company not getting its way in the US and a couple of other countries to boot. And what about the European Union?
Several observations are warranted:
- Tim Cook should be paranoid. Lots of people are out to get Apple and he will be collateral damage.
- What happens if the iPhone craters? Will Apple TV blossom or blow?
- How many pro-Apple humans will suffer bouts of depression? My guess? Lots.
Net net: Numerologists will perceive 2025 as a year for Apple to reflect and prepare for new cycles. I just see 2025 as a triangular number with Tim Apple in its perimeter and no way out evident.
Stephen E Arnold, May 30, 2025
Copilot Disappointments: You Are to Blame
May 30, 2025
No AI, just a dinobaby and his itty bitty computer.
Another interesting Microsoft story from a pro-Microsoft online information service. Windows Central published “Microsoft Won’t Take Bigger Copilot Risks — Due to ‘a Post-Traumatic Stress Disorder from Embarrassments,’ Tracing Back to Clippy.” Why not invoke Bob, the US government suggesting Microsoft security was needy, or the software of the Surface Duo?
The write up reports:
Microsoft claims Copilot and ChatGPT are synonymous, but three-quarters of its AI division pay out of pocket for OpenAI’s superior offering because the Redmond giant won’t allow them to expense it.
Is Microsoft saving money or is Microsoft’s cultural momentum maintaining the velocity of Steve Ballmer taking an Apple iPhone from an employee and allegedly stomping on the device. That helped make Microsoft’s management approach clear to some observers.
The Windows Central article adds:
… a separate report suggested that the top complaint about Copilot to Microsoft’s AI division is that “Copilot isn’t as good as ChatGPT.” Microsoft dismissed the claim, attributing it to poor prompt engineering skills.
This statement suggests that Microsoft is blaming a user for the alleged negative reaction to Copilot. Those pesky users again. Users, not Microsoft, is at fault. But what about the Microsoft employees who seem to prefer ChatGPT?
Windows Central stated:
According to some Microsoft insiders, the report details that Satya Nadella’s vision for Microsoft Copilot wasn’t clear. Following the hype surrounding ChatGPT’s launch, Microsoft wanted to hop on the AI train, too.
I thought the problem was the users and their flawed prompts. Could the issue be Microsoft’s management “vision”? I have an idea. Why not delegate product decisions to Copilot. That will show the users that Microsoft has the right approach to smart software: Cutting back on data centers, acquiring other smart software and AI visionaries, and putting Copilot in Notepad.
Stephen E Arnold, May 30, 2025
AI Can Do Your Knowledge Work But You Will Not Lose Your Job. Never!
May 30, 2025
The dinobaby wrote this without smart software. How stupid is that?
Ravical is going to preserve jobs for knowledge workers. Nevertheless, the company’s AI may complete 80% of the work in these types of organizations. No bean counter on earth would figure out that reducing humanoid workers would cut costs, eliminate the useless vacation scam, and chop the totally unnecessary health care plan. None.
The write up “Belgian AI Startup Says It Can Automate 80% of Work at Expert Firms” reports:
Joris Van Der Gucht, Ravical’s CEO and co-founder, said the “virtual employees” could do 80% of the work in these firms. “Ravical’s agents take on the repetitive, time-consuming tasks that slow experts down,” he told TNW, citing examples such as retrieving data from internal systems, checking the latest regulations, or reading long policies. Despite doing up to 80% of the work in these firms, Van Der Gucht downplayed concerns about the agents supplanting humans.
I believe this statement is 100 percent accurate. AI firms do not use excessive statements to explain their systems and methods. The article provides more concrete evidence that this replacement of humans is spot on:
Enrico Mellis, partner at Lakestar, the lead investor in the round, said he was excited to support the company in bringing its “proven” experience in automation to the booming agentic AI market. “Agentic AI is moving from buzzword to board-level priority,” Mellis said.
Several observations:
- Humans absolutely will be replaced, particularly those who cannot sell
- Bean counters will be among the first to point out that software, as long as it is good enough, will reduce costs
- Executives are judged on financial performance, not the quality of the work as long as revenues and profits result.
Will Ravical become the go-to solution for outfits engaged in knowledge work? No, but it will become a company that other agentic AI firms will watch closely. As long as the AI is good enough, humanoids without the ability to close deals will have plenty of time to ponder opportunities in the world of good enough, hallucinating smart software.
Stephen E Arnold, May 30, 2025
-
- Subscribe to Beyond Search
Feature archive
News archive
-
