Big Numbers and Bad Output: Is This the Google AI Story
May 13, 2025
No AI. Just a dinobaby who gets revved up with buzzwords and baloney.
Alphabet Google reported financials that made stakeholders happy. Big numbers were thrown about. I did not know that 1.5 billion people used Google’s AI Overviews. Well, “use” might be misleading. I think the word might be “see” or “were shown” AI Overviews. The key point is that Google is making money despite its legal hassles and its ongoing battle with infrastructure costs.
I was, therefore, very surprised to read “Google’s AI Overviews Explain Made-Up Idioms With Confident Nonsense.” If the information in the write up is accurate, the factoid suggests that a lot of people may be getting bogus information. If true, what does this suggest about Alphabet Google?
The Cnet article says:
…the author and screenwriter Meaghan Wilson Anastasios shared what happened when she searched “peanut butter platform heels.” Google returned a result referencing a (not real) scientific experiment in which peanut butter was used to demonstrate the creation of diamonds under high pressure.
Those Nobel prize winners, brilliant Googlers, and long-time wizards like Jeff Dean seem to struggle with simple things. Remember the glue cheese on pizza suggestion before Google’s AI improved.
The article adds by quoting a non-Google wizard:
“They [large language models] are designed to generate fluent, plausible-sounding responses, even when the input is completely nonsensical,” said Yafang Li, assistant professor at the Fogelman College of Business and Economics at the University of Memphis. “They are not trained to verify the truth. They are trained to complete the sentence.”
Turning in lousy essay and showing up should be enough for a C grade. Is that enough for smart software with 1.5 billion users every three or four weeks?
The article reminds its readers”
This phenomenon is an entertaining example of LLMs’ tendency to make stuff up — what the AI world calls “hallucinating.” When a gen AI model hallucinates, it produces information that sounds like it could be plausible or accurate but isn’t rooted in reality.
The outputs can be amusing for a person able to identify goofiness. But a grade school kid? Cnet wants users to craft better prompts.
I want to be 17 years old again and be a movie star. The reality is that I am 80 and look like a very old toad.
AI has to make money for Google. Other services are looking more appealing without the weight of legal judgments and hassles in numerous jurisdictions. But Google has already won the AI race. Its DeepMind unit is curing disease and crushing computational problems. I know these facts because Google’s PR and marketing machine is running at or near its red line.
But the 1.5 billion users potentially receiving made up, wrong, or hallucinatory information seems less than amusing to me.
Stephen E Arnold, May 13, 2025
China Smart, US Dumb: Twisting the LLM Daozi
May 12, 2025
No AI, just the dinobaby expressing his opinions to Zellenials.
That hard-hitting technology information service Venture Beat published an interesting article. Its title is “Alibaba ZeroSearch Lets AI Learn to Google Itself — Slashing Training Costs by 88 Percent.” The main point of the write up, in my opinion, is that Chinese engineers have done something really “smart.” The knife at the throat of US smart software companies is cost. The money fires will flame out unless more dollars are dumped into the innovation furnaces of smart software.
The Venture Beat story makes the point that “could dramatically reduce the cost and complexity of training AI systems to search for information, eliminating the need for expensive commercial search engine APIs altogether.”
Oh, oh.
This is smart. Buring cash in pursuit of a fractional improvement is dumb, well, actually, stupid, if the write up’s inforamtion is accurate.
The Venture Beat story says:
The technique, called “ZeroSearch,” allows large language models (LLMs) to develop advanced search capabilities through a simulation approach rather than interacting with real search engines during the training process. This innovation could save companies significant API expenses while offering better control over how AI systems learn to retrieve information.
Is this a Snorkel variant hot from Stanford AI lab?
The write up does not delve into the synthetic data short cut to smart software. After some mumbo jumbo, the write up points out the meat of the “innovation”:
The cost savings are substantial. According to the researchers’ analysis, training with approximately 64,000 search queries using Google Search via SerpAPI would cost about $586.70, while using a 14B-parameter simulation LLM on four A100 GPUs costs only $70.80 — an 88% reduction.
Imagine. A dollar in cost becomes $0.12. If accurate, what should a savvy investor do? Pump money into an outfit like OpenAI or the Xai- type entity, or think harder about the China-smart solution?
Venture Beat explains the implication of the alleged cost savings:
The impact could be substantial for the AI industry.
No kidding?
The Venture Beat analysts add this observation:
The irony is clear: in teaching AI to search without search engines, Alibaba may have created a technology that makes traditional search engines less necessary for AI development. As these systems become more self-sufficient, the technology landscape could look very different in just a few years.
Yep, irony. Free transformer technology. Free Snorkle technology. Free kinetic into the core of the LLM money furnace.
If true, the implications are easy to outline. If bogus, the China Smart, US Dumb trope still captured ink and will be embedded in some smart software’s increasingly frequent hallucinatory outputs. At which point, the China Smart, US Dumb information gains traction and becomes “fact” to some.
Stephen E Arnold, May 12, 2025
Another Duh! Moment: AI Cannot Read Social Situations
May 12, 2025
No AI. Just a dinobaby who gets revved up with buzzwords and baloney.
I promise I won’t write “Duh!” in this blog post again. I read Science Daily’s story “Awkward. Humans Are Still Better Than AI at Reading the Room.” The write up says without total awareness:
Humans, it turns out, are better than current AI models at describing and interpreting social interactions in a moving scene — a skill necessary for self-driving cars, assistive robots, and other technologies that rely on AI systems to navigate the real world.
Yeah, what about in smart weapons, deciding about health care for an elderly patient, or figuring out whether the obstacle is a painted barrier designed to demonstrate that full self driving is a work in progress. (I won’t position myself in front of a car with auto-sensing and automatic braking. You can have at it.)
The write up adds:
Video models were unable to accurately describe what people were doing in the videos. Even image models that were given a series of still frames to analyze could not reliably predict whether people were communicating. Language models were better at predicting human behavior, while video models were better at predicting neural activity in the brain.
Do these findings say to you, “Not ready for prime time?” It does to me.
One of the researchers who was in the weeds with the data points out:
“I think there’s something fundamental about the way humans are processing scenes that these models are missing.”
Okay, I prevaricated. Duh!” (Do marketers care? Duh!)
Stephen E Arnold, May 12, 2025
Google, Its AI Search, and Web Site Traffic
May 12, 2025
No AI. Just a dinobaby sharing an observation about younger managers and their innocence.
I read “Google’s AI Search Switch Leaves Indie Websites Unmoored.” I think this is a Gen Y way of saying, “No traffic for you, bozos.” Of course, as a dinobaby, I am probably wrong.
Let’s look at the write up. It says:
many publishers said they either need to shut down or revamp their distribution strategy. Experts this effort could ultimately reduce the quality of information Google can access for its search results and AI answers.
Okay, but this is just one way to look at Google’s delicious decision.
May I share some of my personal thoughts about what this traffic downshift means for those blue-chip consultant Googlers in charge:
First, in the good old days before the decline began in 2006, Google indexed bluebirds (sites that had to be checked for new content or “deltas” on an accelerated heart beat. Examples were whitehouse.gov (no, not the whitehouse.com porn site). Then there were sparrows. These plentiful Web sites could be checked on a relaxed schedule. I mean how often do you visit the US government’s National Railway Retirement Web site if it still is maintained and online? Yep, the correct answer is, “Never.” There there were canaries. These were sites which might signal a surge in popularity. They were checked on a heart beat that ensured the Google wouldn’t miss a trend and fail to sell advertising to those lucky ad buyers.
So, bluebirds, canaries, and sparrows.
This shift means that Google can reduce costs by focusing on bluebirds and canaries. The sparrows — the site operated by someone’s grandmother to sell home made quilts — won’t get traffic unless the site operator buys advertising. It’s pay to play. If a site is not in the Google index, it just may not exist. Sure there are alternative Web search systems, but none, as far as I know, are close to the scope of the “old” Google in 2006.
Second, by dropping sparrows or pinging them once in a blue moon will reduce the costs of crawling, indexing, and doing the behind-the-scenes work that consumes Google cash at an astonishing rate. Therefore, the myth of indexing the “Web” is going to persist, but the content of the index is not going to be “fresh.” This is the concept that some sites like whitehouse.gov have important information that must be in search results. Non-priority sites just disappear or fade. Eventually the users won’t know something is missing, which is assisted by the decline in education for some Google users. The top one percent knows bad or missing information. The other 99 percent? Well, good luck.
Third, the change means that publishers will have some options. [a] They can block Google’s spider and chase the options. How’s Yandex.ru sound? [b] They can buy advertising and move forward. I suggest these publishers ask a Google advertising representative what the minimum spend is to get traffic. [c] Publishers can join together and try to come up with a joint effort to resist the increasingly aggressive business actions of Google. Do you have a Google button on your remote? Well, you will. [d] Be innovative. Yeah, no comment.
Net net: This item about the impact of AI Overviews is important. Just consider what Google gains and the pickle publishers and other Web sites now find themselves enjoying.
Stephen E Arnold, May 12, 2025
Microsoft AI: Little Numbers and Mucho Marketing
May 10, 2025
No AI. Just a dinobaby who gets revved up with buzzwords and baloney.
I am confused. The big AI outfits have spent and are spending big bucks on [a] marketing, [b] data centers, [c] marketing, [d] chips, [e] reorganizations, and [f] marketing. I think I have the main cost centers, but I may have missed one. Yeah, I did. Marketing.
Has the AI super machine run into some problems? Thanks, MidJourney, you were online today unlike OpenAI.
What is AI doing? It is definitely selling consulting services. Some wizards are using it to summarize documents because that takes a human time to do. You know: Reading, taking notes, then capturing the juicy bits. Let AI do that. And customer support? Yes, much cheaper some say than paying humans to talk to a mere customer.
Imagine my surprise when I read “Microsoft’s Big AI Hire Can’t Match OpenAI.” Microsoft’s AI leader among AI leaders, according to the write up, “hasn’t delivered the turnaround he was hired to bring.” Microsoft caught the Google by surprise a couple of years ago, caused a Googley Code Red or Yellow or whatever, and helped launch the “AI is the next big thing innovators have been waiting for.”
The write up asserts:
At Microsoft’s annual executive huddle last month, the company’s chief financial officer, Amy Hood, put up a slide that charted the number of users for its Copilot consumer AI tool over the past year. It was essentially a flat line, showing around 20 million weekly users. On the same slide was another line showing ChatGPT’s growth over the same period, arching ever upward toward 400 million weekly users. OpenAI’s iconic chatbot was soaring, while Microsoft’s best hope for a mass-adoption AI tool was idling.
Keep in mind that Google suggested it had 1.5 billion users of its Gemini service, and (I think) Google implied that its AI is the quantumly supreme smart software. I may have that wrong and Google’s approach just wins chess, creates new drugs, and suggests that one can glue cheese on pizza. I may have these achievements confused, but I am an 80 year old dinobaby and easily confused these days.
The write up also contains some information I found a bit troubling; to wit:
And at this point, Microsoft is just not in the running to build a model that can compete with the best from OpenAI, Anthropic, Google, and even xAI. The projects that people have mentioned to me feel incremental, as opposed to leapfrogging the competition.
One can argue that Microsoft does not have to be in the big leagues. The company usually takes three or more releases to get a feature working. (How about those printers that don’t work?) The number of Softie software users is big. Put the new functionality in a release and — bingo! — market leadership. That SharePoint is a wonderful content management system. Just ask some of the security team in the Israeli military struggling with a “squadron” fumble.
Several observations:
- Microsoft’s dialing back some data center action may be a response to the under performance of its AI is the future push. If not, then maybe Microsoft has just pulled a Bob or a Clippy?
- I am not sure that the payoffs for other AI leaders’ investments are going to grab the brass ring or produce a winning lottery ticket. So many people desperately want AI to deliver dump trucks of gold dust to their cubicles that the neediness is palpable. AI is — it must be — the next big thing.
- Users are finding that for some use cases, AI is definitely a winner. College students use it to make more free time for hanging out and using TikTok-type services. Law firms find that AI is good enough to track down obscure cases that can be used in court as long as a human who knows the legal landscape checks the references before handing them over to a judge who can use an ATM machine and a mobile phone for voice calls. For many applications, the hallucination issue looms large.
- China’s free smart software models work reasonably well and have ignited such diverse applications as automated pig butchering and proving that cheaper CPUs and GPUs work in a “good enough” way just for less money.
I don’t want to pick on Microsoft, but I want to ask a question, “Is this the first of the big technology companies hungry and thirsty for the next big thing starting to find out that AI may not deliver?”
Stephen E Arnold, May 13, 2025
The Future: Humans in Lawn Chairs. Robots Do the Sports Thing
May 8, 2025
Can a fast robot outrun a fast human? Not yet, apparently. MSN’s Interesting Engineering reports, “Humanoid ‘Tiangong Ultra’ Dons Winning Boot in World’s First Human Vs Robot Marathon.” In what appears to be the first event of its kind, a recent 13-mile marathon pitted robots and humans against each other in Beijing. Writer Christopher McFadden reports:
“Around 21 humanoid robots officially competed alongside human marathoners in a 13-mile (21 km) endurance race in Beijing on Saturday, April 19th. According to reports, this is the first time such an event has been held. Competitor robots varied in size, with some as short as 3 feet 9 inches (1.19 m) and others as tall as 5 feet 9 inches (1.8 m). Wheeled robots were officially banned from the race, necessitating that any entrants be able to walk or run similarly to humans.”
The winner was one of the tallest at 5 feet 9 inches and weighed 114 pounds. It took Tiangong Ultra two hours and forty minutes to complete the course. Despite its impressive performance, it lagged considerably behind the first-place human who finished at one hour and two minutes. The robots’ lane of the course was designed to test the machines’ capabilities, mixing inclines and both left and right turns with flat stretches.
See the article for a short video of the race. Most of it features the winner, but there is a brief shot of one smaller, cuter robot. The article continues:
“According to the robot’s creator, Tang Jian, who is also the chief technology officer behind the Beijing Innovation Centre of Human Robotics, the robot’s long legs and onboard software both aided it in its impressive feat. … Jian added that the robot’s battery needed to be changed only three times during the race. As for other robot entrants, many didn’t perform as well. In particular, one robot fell at the starting line and lay on the ground for a few minutes before getting up and joining the race. Yet another crashed into a railing, causing its human operator to fall over.”
Oops. Sadly, those incidents do not appear in the video. The future is clear: Wizards will sit in lawn chairs and watch their robots play sports. I wonder if my robot will go to the gym and exercise for me?
Cynthia Murrell, May 8, 2025
IBM: Making the Mainframe Cool Again
May 7, 2025
No AI, just the dinobaby expressing his opinions to Zellenials.
I a ZDNet Tech Today article titled “IBM Introduces a Mainframe for AI: The LinuxONE Emperor 5.” Years ago, I had three IBM PC 704s, each with the eight drive SCSI chassis and that wonderful ServeRAID software. I suppose I should tell you, I want a LinuxONE Emperor 5 because the capitalization reminds me of the IBM ServeRAID software. Imagine. A mainframe for artificial intelligence. No wonder that IBM stock looks like a winner in 2025.
The write up says:
IBM’s latest mainframe, the LinuxONE Emperor 5, is not your grandpa’s mainframe
The CPU for this puppy is the IBM Telum II processor. The chip is a seven nanometer item announced in 2021. If you want some information about this, navigate to “IBM’s Newest Chip Is More Than Meets the AI.”
The ZDNet write up says:
Manufactured using Samsung’s 5 nm process technology, Telum II features eight high-performance cores running at 5.5GHz, a 40% increase in on-chip cache capacity (with virtual L3 and L4 caches expanded to 360MB and 2.88GB, respectively), and a dedicated, next-generation on-chip AI accelerator capable of up to 24 trillion operations per second (TOPS) — four times the compute power of its predecessor. The new mainframe also supports the IBM Spyre Accelerator for AI users who want the most power.
The ZDNet write up delivers a bumper crop of IBM buzzwords about security, but there is one question that crossed my mind, “What makes this a mainframe?”
The answer, in my opinion, is IBM marketing. The Emperor should be able to run legacy IBM mainframe applications. However, before placing an order, a customer may want to consider:
- Snapping these machines into a modern cloud or hybrid environment might take a bit of work. Never fear, however, IBM consulting can help with this task.
- The reliance on the Telum CPU to do AI might put the system at a performance disadvantage from solutions like the Nvidia approach
- The security pitch is accurate providing the system is properly configured and set up. Once again, IBM provides the for fee services necessary to allow Z-llenial IT professional to sleep easy on weekends.
- Mainframes in the cloud are time sharing oriented; making these work in a hybrid environment can be an interesting technical challenge. Remember: IBM consulting and engineering services can smooth the bumps in the road.
Net net: Interesting system, surprising marketing, and definitely something that will catch a bean counter’s eye.
Stephen E Arnold, May 7, 2025
Microsoft Explains that Its AI Leads to Smart Software Capacity Gap Closing
May 7, 2025
No AI, just a dinobaby watching the world respond to the tech bros.
I read a content marketing write up with two interesting features: [1] New jargon about smart software and [2] a direct response to Google’s increasingly urgent suggestions that Googzilla has won the AI war. The article appears in Venture Beat with the title “Microsoft Just Launched Powerful AI ‘Agents’ That Could Completely Transform Your Workday — And Challenge Google’s Workplace Dominance.” The title suggests that Google is the leader in smart software in the lucrative enterprise market. But isn’t Microsoft’s “flavor” of smart software in products from the much-loved Teams to the lowly Notepad application? Isn’t Word like Excel at the top of the heap when it comes to usage in the enterprise?
I will ignore these questions and focus on the lingo in the article. It is different and illustrates what college graduates with a B.A. in modern fiction can craft when assisted by a sprinkling of social science majors and a former journalist or two.
Here are the terms I circled:
product name: Microsoft 365 Copilot Wave 2 Spring release (wow, snappy)
integral collaborator (another bound phrase which means agent)
intelligence abundance (something everyone is talking about)
frontier firm (forward leaning synonym)
‘human-led, agent-operated’ workplaces (yes, humans lead; they are not completely eliminated)
agent store (yes, another online store. You buy agents; you don’t buy people)
browser for AI
brand-compliant images
capacity gap (I have no idea what this represents)
agent boss (Is this a Copilot thing?)
work charts (not images, plans I think)
Copilot control system (Is this the agent boss thing?)
So what does the write up say? In my dinobaby mind, the answer is, “Everything a member of leadership could want: Fewer employees, more productivity from those who remain on the payroll, software middle managers who don’t complain or demand emotional support from their bosses, and a narrowing of the capacity gap (whatever that is).
The question is, “Can either Google, Microsoft, or OpenAI deliver this type of grand vision?” Answer: Probably the vision can be explained and made magnetic via marketing, PR, and language weaponization, but the current AI technology still has a couple of hurdles to get over without tearing the competitors’ gym shorts:
- Hallucinations and making stuff up
- Copyright issues related to training and slapping the circle C, trademarks, and patents on outputs from these agent bosses and robot workers
- Working without creating a larger attack surface for bad actors armed with AI to exploit (Remember, security, not AI, is supposed to be Job No. 1 at Microsoft. You remember that, right? Right?)
- Killing dolphins, bleaching coral, and choking humans on power plant outputs
- Getting the billions pumped into smart software back in the form of sustainable and growing revenues. (Yes, there is a Santa Claus too.)
Net net: Wow. Your turn Google. Tell us you have won, cured disease, and crushed another game player. Oh, you will have to use another word for “dominance.” Tip: Let OpenAI suggest some synonyms.
Stephen E Arnold, May 7, 2025
Google Versus OpenAI: Whose Fish Is Bigger?
May 6, 2025
No AI, just a dinobaby watching the world respond to the tech bros.
Bing Crosby quipped on one of his long-ago radio shows, “We are talking about fish here” when asked about being pulled to shore by a salmon he caught. I think about the Bingster when I come across “user” numbers for different smart software systems. “Google Reveals Sky High Gemini Usage Numbers in Antitrust Case” provides some perjury proof data that it is definitely number two in smart software.
According to the write up:
The [Google] slide listed Gemini’s 350 million monthly users, along with daily traffic of 35 million users.
Okay, we have some numbers.
The write up provides a comparative set of data; to wit:
OpenAI has also seen traffic increase, putting ChatGPT around 600 million monthly active users, according to Google’s analysis. Early this year, reports pegged ChatGPT usage at around 400 million users per month.
Where’s Microsoft in this count? Yeah, who knows? MSFT just pounds home that it is winning in the enterprise. Okay, I understand.
What’s interesting about these data or lack of it has several facets:
- For Google, the “we’re number two” angle makes clear that its monopoly in online advertising has not transferred to becoming automatically number one in AI
- The data from Google are difficult to verify, but everyone trusts the Google
- The data from OpenAI are difficult to verify, but everyone trusts Sam AI-Man.
Where are we in the AI game?
At the mercy of unverifiable numbers and marketing type assertions.
What about Deepseek which may be banned by some of the folks in Washington, DC? What about everyone’s favorite litigant Meta / Facebook?
Net net: AI is everywhere so what’s the big deal? Let’s get used to marketing because those wonderful large language models still have a bit of problem with hallucinations, not to mention security issues and copyright hassles. I won’t mention cost because the data make clear that the billions pumped into smart software have not generated a return yet. Someday perhaps?
Stephen E Arnold, May 6, 2025
Deep Fake Recognition: Google Has a Finger In
May 5, 2025
Sorry, no AI used to create this item.
I spotted this Newsweek story: “‘AI Imposter’ Candidate Discovered During Job Interview, Recruiter Warns.” The main idea is that a humanoid struggled to identify a deep fake. The deep fake was applying for a job.
The write up says:
Several weeks ago, Bettina Liporazzi, the recruiting lead at letsmake.com was contacted by a seemingly ordinary candidate who was looking for a job. Their initial message was clearly AI-generated, but Liporazzi told Newsweek that this “didn’t immediately raise any flags” because that’s increasingly commonplace.
Here’s the interesting point:
Each time the candidate joined the call, Liporazzi got a warning from Google to say the person wasn’t signed in and “might not be who they claim to be.”
This interaction seems to have taken place online.
The Newsweek story includes this statement:
As generative-AI becomes increasingly powerful, the line between what’s real and fake is becoming harder to decipher. Ben Colman, co-founder and CEO of Reality Defender, a deepfake detection company, tells Newsweek that AI impersonation in recruiting is “just the tip of the iceberg.”
The recruiter figured out something was amiss. However, in the sequence Google injected its warning.
Several questions:
- Does Google monitor this recruiter’s online interactions and analyze them?
- How does Google determine which online interaction is one in which it should simply monitor and which to interfere?
- What does Google do with the information about [a] the recruiter, [b] the job on offer itself, and [c] the deep fake system’s operator?
I wonder if Newsweek missed the more important angle in this allegedly actual factual story; that is, Google surveillance? Perhaps Google was just monitoring email when it tells me that a message from a US law enforcement agency is not in my list of contacts. How helpful, Google?
Will Google’s “monitoring” protect others from Deep Fakes? Those helpful YouTube notices are part of this effort to protect it seems.
Stephen E Arnold, May 5, 2025
-
- Subscribe to Beyond Search
Feature archive
News archive
-
