The AI Profit and Cost Race: Drivers, Get Your Checkbooks Out
January 15, 2025
A dinobaby-crafted post. I confess. I used smart software to create the heart wrenching scene of a farmer facing a tough 2025.
Microsoft appears ready to spend $80 billion “on AI-enabled data centers” by December 31, 2025. Half of the money will go to US facilities, and the other half, other nation states. I learned this information from a US cable news outfit’s article “Microsoft Expects to Spend $80 Billion on AI-Enabled Data Centers in Fiscal 2025.” Is Microsoft tossing out numbers as part of a marketing plan to trigger the lustrous Google, or is Microsoft making clear that it is going whole hog for smart software despite the worries of investors that an AI revenue drought persists? My thought is that Microsoft likes to destabilize the McKinsey-type thinking at Google, wait for the online advertising giant to deliver another fabulous Sundar & Prabhakar Comedy Tour, and then continue plodding forward.
The write up reports:
Several top-tier technology companies are rushing to spend billions on Nvidia graphics processing units for training and running AI models. The fast spread of OpenAI’s ChatGPT assistant, which launched in late 2022, kicked off the AI race for companies to deliver their own generative AI capabilities. Having invested more than $13 billion in OpenAI, Microsoft provides cloud infrastructure to the startup and has incorporated its models into Windows, Teams and other products.
Yep, Google-centric marketing.
Thanks, You.com. Good enough.
But if Microsoft does spend $80 billion, how will the company convert those costs into a profit geyser? That’s a good question. Microsoft appears to be cooperating with discounts for its mainstream consumer software. I saw advertisements offering Windows 11 Professional for $25. Other deep discounts can be found for Office 365, Visio, and even the bread-and-butter sales pitch PowerPoint application.
Tweaking Google is one thing. Dealing with cost competition is another.
I noted that the South China Morning Post’s article “Alibaba Ties Up with Lee Kai-fu’s Unicorn As China’s AI Sector Consolidates.” Tucked into that rah rah write up was this statement:
The cooperation between two of China’s top AI players comes as price wars continue in the domestic market, forcing companies to further slash prices or seek partnerships with former foes. Alibaba Cloud said on Tuesday it would reduce the fees for using its visual reasoning AI model by up to 85 per cent, the third time it had marked down the prices of its AI services in the past year. That came after TikTok parent ByteDance last month cut the price of its visual model to 0.003 yuan (US$0.0004) per thousand token uses, about 85 per cent lower than the industry average.
The message is clear. The same tactic that China’s electric vehicle manufacturers are using will be applied to smart software. The idea is that people will buy good enough products and services if the price is attractive. Bean counters intuitively know that a competitor that reduces prices and delivers an acceptable product can gain market share. The companies unable to compete on price face rising costs and may be forced to cut their prices, thus risking financial collapse.
For a multi-national company, the cost of Chinese smart software may be sufficiently good to attract business. Some companies which operate under US sanctions and controls of one type or another may be faced with losing some significant markets. Examples include Brazil, India, Middle Eastern nations, and others. That means that a price war can poke holes in the financial predictions which outfits like Microsoft are basing some business decisions.
What’s interesting is that this smart software tactic apparently operating in China fits in with other efforts to undermine some US methods of dominating the world’s financial system. I have no illusions about the maturity of the AI software. I am, however, realistic about the impact of spending significant sums with the fervent belief that a golden goose will land on the front lawn of Microsoft’s headquarters. I am okay with talking about AI in order to wind up Google. I am a bit skeptical about hosing $80 billion into data centers. These puppies gobble up power, which is going to get expensive quickly if demand continues to blast past the power generation industry’s projections. An economic downturn in 2025 will not help ameliorate the situation. Toss in regional wars and social turmoil and what does one get?
Risk. Welcome to 2025.
Stephen E Arnold, January 15, 2025
AI Search Engine from Alibaba Grows Apace
January 15, 2025
Prepared by a still-alive dinobaby.
The Deepseek red herring has been dragged across the path of US AI innovators. A flurry of technology services wrote about Deepseek’s ability to give US smart software companies a bit of an open source challenge. The hook, however, was not just the efficacy of the approach. The killer message was, “Better, faster, and cheaper.” Yep, cheaper, the concept which raises questions about certain US outfits burning cash in units of a one billion dollars with every clock tick.
A number of friendly and lovable deer are eating the plants in Uncle Sam’s garden. How many of these are living in the woods looking for a market to consume? Thanks OpenAI, good enough.
Now Alibaba is coming for AI search. The Chinese company crows on PR Newswire, "Alibaba’s Accio AI Search Engine Hits 500,000 SME User Milestone." Sounds like a great solution for US businesses doing work for the government. The press release reveals:
"Alibaba International proudly announces that its artificial intelligence (AI)-powered business-to-business (B2B) search engine for product sourcing, Accio, has reached a significant milestone since its launch in November 2024, currently boasting over 500,000 small and medium-sized enterprise (SME) users. … During the peak global e-commerce sales seasons in November and December, more than 50,000 SMEs worldwide have actively used Accio to source inspirations for Black Friday and Christmas inventory stocking. User feedback shows that the search engine helped them achieve this efficiently. Accio now holds a net promoter score (NPS) exceeding 50[1], indicating a high level of customer satisfaction. On December 13, 2024, the dynamic search engine was also named ‘Product of the Day’ on Product Hunt, a site that curates new products in tech, further cementing its status as an indispensable tool for SME buyers worldwide."
Well, good for them. And, presumably, for China ‘s information gathering program. Founded in 1999, Alibaba Group is based in Hangzhou, Zhejiang. One can ask many questions about Alibaba, including ones related to the company’s interaction with Chinese government officials. When a couple of deer are eating one’s garden vegetables, a good question to ask is, “How many of these adorable creatures live in the woods?” One does not have to be Natty Bumpo to know that the answer is, “There are more where those came from.”
Cynthia Murrell, January 15, 2025
Agentic Workflows and the Dust Up Between Microsoft and Salesforce
January 14, 2025
Prepared by a still-alive dinobaby.
The Register, a UK online publication, does a good job of presenting newsworthy events with a touch of humor. Today I spotted a new type of information in the form of an explainer plus management analysis. Plus the lingo and organization suggest a human did all or most of the work required to crank out a very good article called “In AI Agent Push, Microsoft Re-Orgs to Create CoreAI – Platform and Tools Team.”
I want to highlight the explainer part of the article. The focus is on the notion of agentic; specifically:
agentic applications with memory, entitlements, and action space that will inherit powerful model capabilities. And we will adapt these capabilities for enhanced performance and safety across roles, business processes, and industry domains. Further, how we build, deploy, and maintain code for these AI applications is also fundamentally changing and becoming agentic.
These words are attributed to Microsoft’s top dog Satya Nadella, but they sound as if one of the highly paid wordsmiths laboring for the capable Softies. Nevertheless, the idea is important. In order to achieve the agentic pinnacle, Microsoft has to reorganize. Whoever can figure out how to make agentic applications work across different vendors’ solutions will be able to make money. That’s the basic idea: Smart software is going to create a new big thing for enterprise software and probably some consumers.
The write up explains:
It’s arguably just plain old software talking to plain old software, which would be nothing new. The new angle here, though, is that it’s driven mainly by, shall we say, imaginative neural networks and models making decisions, rather than algorithms following entirely deterministic routes. Which is still software working with software. Nadella thinks building artificially intelligent agentic apps and workflows needs “a new AI-first app stack — one with new UI/UX patterns, runtimes to build with agents, orchestrate multiple agents, and a reimagined management and observability layer.”
To win the land in this new territory, Microsoft must have a Core AI team. Google and Salesforce presumably have this type of set up. Microsoft has to step up its AI efforts. The Register points out:
Nadella noted that “our internal organizational boundaries are meaningless to both our customers and to our competitors”. That’s an odd observation given Microsoft published his letter, which concludes with this observation: “Our success in this next phase will be determined by having the best AI platform, tools, and infrastructure. We have a lot of work to do and a tremendous opportunity ahead, and together, I’m looking forward to building what comes next.”
Here’s what I found interesting:
- Agentic is the next big thing in smart software. Essentially smart software that does one thing is useful. Orchestrating agents to do a complex process is the future. The software decides. Everything works well — at least, that’s the assumption.
- Microsoft, like Google, is now in a Code Yellow or Code Red mode. The company feels the heat from Salesforce. My hunch is that Microsoft knows that add ins like Ghostwriter for Microsoft Office is more useful than Microsoft’s own Copilot for many users. If the same boiled fish appears on the enterprise menu, Microsoft is in a world of hurt from Salesforce and probably a lot of other outfits.
- The re-org parallels the disorder that surfaced at Google when it fixed up its smart software operation or tried to deal with the clash of the wizards in that estimable company. Pushing boxes around on an organization chart is honorable work, but that management method may not deliver the agentic integration some people want.
The conclusion I drew from The Register’s article is that the big AI push and the big players’ need to pop up a conceptual level in smart software is perceived as urgent. Costs? No problem. Hallucination? No problem. Hardware availability? No problem. Software? No problem. A re-organization is obvious and easy. No problem.
Stephen E Arnold, January 14, 2025
More about NAMER, the Bitext Smart Entity Technology
January 14, 2025
A dinobaby product! We used some smart software to fix up the grammar. The system mostly worked. Surprised? We were.
We spotted more information about the Madrid, Spain based Bitext technology firm. The company posted “Integrating Bitext NAMER with LLMs” in late December 2024. At about the same time, government authorities arrested a person known as “Broken Tooth.” In 2021, an alert for this individual was posted. His “real” name is Wan Kuok-koi, and he has been in an out of trouble for a number of years. He is alleged to be part of a criminal organization and active in a number of illegal behaviors; for example, money laundering and human trafficking. The online service Irrawady reported that Broken Tooth is “the face of Chinese investment in Myanmar.”
Broken Tooth (né Wan Kuok-koi, born in Macau) is one example of the importance of identifying entity names and relating them to individuals and the organizations with which they are affiliated. A failure to identify entities correctly can mean the difference between resolving an alleged criminal activity and a get-out-of-jail-free card. This is the specific problem that Bitext’s NAMER system addresses. Bitext says that large language models are designed for for text generation, not entity classification. Furthermore, LLMs pose some cost and computational demands which can pose problems to some organizations working within tight budget constraints. Plus, processing certain data in a cloud increases privacy and security risks.
Bitext’s solution provides an alternative way to achieve fine-grained entity identification, extraction, and tagging. Bitext’s solution combines classical natural language processing solutions solutions with large language models. Classical NLP tools, often deployable locally, complement LLMs to enhance NER performance.
NAMER excels at:
- Identifying generic names and classifying them as people, places, or organizations.
- Resolving aliases and pseudonyms.
- Differentiating similar names tied to unrelated entities.
Bitext supports over 20 languages, with additional options available on request. How does the hybrid approach function? There are two effective integration methods for Bitext NAMER with LLMs like GPT or Llama are. The first is pre-processing input. This means that entities are annotated before passing the text to the LLM, ideal for connecting entities to knowledge graphs in large systems. The second is to configure the LLM to call NAMER dynamically.
The output of the Bitext system can generate tagged entity lists and metadata for content libraries or dictionary applications. The NAMER output can integrate directly into existing controlled vocabularies, indexes, or knowledge graphs. Also, NAMER makes it possible to maintain separate files of entities for on-demand access by analysts, investigators, or other text analytics software.
By grouping name variants, Bitext NAMER streamlines search queries, enhancing document retrieval and linking entities to knowledge graphs. This creates a tailored “semantic layer” that enriches organizational systems with precision and efficiency.
For more information about the unique NAMER system, contact Bitext via the firm’s Web site at www.bitext.com.
Stephen E Arnold, January 14, 2025
Some AI Wisdom: Is There a T Shirt?
January 14, 2025
Prepared by a still-alive dinobaby.
I was catching up with newsfeeds and busy filtering the human output from the smart software spam-arator. I spotted “The Serious Science of Trolling LLMs,” published in the summer of 2024. The article explains that value can be derived from testing large language models like ChatGPT, Gemini, and others with prompts to force the software to generate something really stupid, off base, incorrect, or goofy. I zipped through the write up and found it interesting. Then I came upon this passage:
the LLM business is to some extent predicated on deception; we are not supposed to know where the magic ends and where cheap tricks begin. The vendors’ hope is that with time, we will reach full human-LLM parity; and until then, it’s OK to fudge it a bit. From this perspective, the viral examples that make it patently clear that the models don’t reason like humans are not just PR annoyances; they are a threat to product strategy.
Several observations:
- Progress from my point of view with smart software seems to have slowed. The reason may be that free and low cost services cannot affords to provide the functionality they did before someone figured out the cost per query. The bean counters spoke and “quality” went out the window.
- The gap between what the marketers say and what the systems do is getting wider. Sorry, AI wizards, the systems typically fail to produce an output satisfactory for my purposes on the first try. Multiple prompts are required. Again a cost cutting move in my opinion.
- Made up information or dead wrong information is becoming more evident. My hunch is that the consequence of ingesting content produced by AI is degrading the value of the models originally trained on human generated content. I think this is called garbage in — garbage out.
Net net: Which of the deep pocket people will be the first to step back from smart software built upon systems that consume billions of dollars the way my French bulldog eats doggie treats? The Chinese system Deepseek’s marketing essentially says, “Yo, we built this LLM at a fraction of the cost of the spendthrifts at Google, Microsoft, and OpenAI. Are the Chinese AI wizards dragging a red herring around the AI forest?
To go back to the Lcamtuf essay, “it’s OK to fudge a bit.” Nope, it is mandatory to fudge a lot.
Stephen E Arnold, January 14, 2025
AI Defined in an Arts and Crafts Setting No Less
January 13, 2025
Prepared by a still-alive dinobaby.
I was surprised to learn that a design online service (what I call arts and crafts) tackled a to most online publications skip. The article “What Does AI Really Mean?” tries to define AI or smart software. I remember a somewhat confused and erratic college professor trying to define happiness. Wow, that was a wild and crazy lecture. (I think the person’s name was Dr. Chapman. I tip my ball cap with the SS8 logo on it to him.) The author of this essay is a Googler, so it must be outstanding, furthering the notion of quantum supremacy at Google.
What is AI? The write up says:
I hope this helped you better understand what those terms mean and the processes which encompass the term “AI”.
Okay, “helped you understand better.” What does the essay do to help me understand better. Hang on to your SS8 ball cap. The author briefly defines these buzzwords:
- Data as coordinates
- Querying per approximation
- Language models both large and small
- Fine “Tunning” (Isn’t that supposed to be tuning?)
- Enhancing context with information, including grounded generation
- Embedding.
For me, a list of buzzwords is not a definition. (At least the hapless Dr. Chapman tried to provide concrete examples and references to his own experience with happiness, which as I recall eluded him.)
The “definition” jumps to a section called “Let’s build.” The author concludes the essay with:
I hope this helped you better understand what those terms mean and the processes which encompass the term “AI”. This merely scratches the surface of complexity, though. We still need to talk about AI Agents and how all these approaches intertwine to create richer experiences. Perhaps we can do that in a later article — let me know in the comments if you’d like that!
That’s it. The Google has, from his point of view, defined AI. As Holden Caufield in The Catcher in the Rye said:
“I can’t explain what I mean. And even if I could, I’m not sure I’d feel like it.”
Bingo.
Stephen E Arnold, January 13, 2025
Oh, Oh! Silicon Valley Hype Minimizes Risk. Who Knew?
January 10, 2025
This is an official dinobaby post. No smart software involved in this blog post.
I read “Silicon Valley Stifled the AI Doom Movement in 2024.” I must admit I was surprised that one of the cheerleaders for Silicon Valley is disclosing something absolutely no one knew. I mean unregulated monopolies, the “Puff the Magic Dragon” strafing teens, and the vulture capitalists slavering over the corpses of once thriving small and mid sized businesses. Hey, I thought that “progress” myth was real. I thought technology only makes life better. Now I read that “Silicon Valley” wanted only good news about smart software. Keep in mind that this is software which outputs hallucinations, makes decisions about medical care for people, and monitors the clicks and location of everyone with a mobile device or a geotracker.
The write up reminded me that ace entrepreneur / venture professional Marc Andreessen said:
“The era of Artificial Intelligence is here, and boy are people freaking out. Fortunately, I am here to bring the good news: AI will not destroy the world, and in fact may save it,” said Andreessen in the essay. In his conclusion, Andreessen gave a convenient solution to our AI fears: move fast and break things – basically the same ideology that has defined every other 21st century technology (and their attendant problems). He argued that Big Tech companies and startups should be allowed to build AI as fast and aggressively as possible, with few to no regulatory barriers. This would ensure AI does not fall into the hands of a few powerful companies or governments, and would allow America to compete effectively with China, he said.
What publications touted Mr. Andreessen’s vision? Answer: Lots.
Regulate smart software? Nope. From Connecticut’s effort to the US government, smart software regulation went nowhere. The reasons included, in my opinion:
- A chance to make a buck, well, lots of bucks
- Opportunities to foist “smart software” plus its inherent ability to make up stuff on corporate sheep
- A desire to reinvent “dumb” processes like figuring out how to push buttons to create addiction to online gambling, reduce costs by eliminating inefficient humans, and using stupid weapons.
Where are we now? A pillar of the Silicon Valley media ecosystem writes about the possible manipulation of information to make smart software into a Care Bear. Cuddly. Harmless. Squeezable. Yummy too.
The write up concludes without one hint of the contrast between the AI hype and the viewpoints of people who think that the technology of AI is immature but fumbling forward to stick its baby finger in a wall socket. I noted this concluding statement in the write up:
Calling AI “tremendously safe” and attempts to regulate it “dumb” is something of an oversimplification. For example, Character.AI – a startup a16z has invested in – is currently being sued and investigated over child safety concerns. In one active lawsuit, a 14-year-old Florida boy killed himself after allegedly confiding his suicidal thoughts to a Character.AI chatbot that he had romantic and sexual chats with. This case shows how our society has to prepare for new types of risks around AI that may have sounded ridiculous just a few years ago. There are more bills floating around that address long-term AI risk – including one just introduced at the federal level by Senator Mitt Romney. But now, it seems AI doomers will be fighting an uphill battle in 2025.
But don’t worry. Open source AI provides a level playing field for [a] adversaries of the US, [b] bad actors who use smart software to compromise Swiss cheese systems, and [c] manipulate people on a grand scale. Will the “Silicon Valley” media give equal time to those who don’t see technology as a benign or net positive? Are you kidding? Oh, aren’t those smart drones with kinetic devices just fantastic?
Stephen E Arnold, January 10, 2025
GitHub Identifies a Sooty Pot and Does Not Offer a Fix
January 9, 2025
This is an official dinobaby post. No smart software involved in this blog post.
GitLab’s Sabrina Farmer is a sharp thinking person. Her “Three Software Development Challenges Slowing AI Progress” articulates an issue often ignored or just unknown. Specifically, according to her:
AI is becoming an increasingly critical component in software development. However, as is the case when implementing any new tool, there are potential growing pains that may make the transition to AI-powered software development more challenging.
Ms. Farmer is being kind and polite. I think she is suggesting that the nest with the AI eggs from the fund-raising golden goose has become untidy. Perhaps, I should use the word “unseemly”?
She points out three challenges which I interpret as the equivalent of one of those unsolved math problems like cracking the Riemann Hypothesis or the Poincaré Conjecture. These are:
- AI training. Yeah, marketers write about smart software. But a relatively small number of people fiddle with the knobs and dials on the training methods and the rat’s nests of computational layers that make life easy for an eighth grader writing an essay about Washington’s alleged crossing of the Delaware River whilst standing up in a boat rowed by hearty, cheerful lads. Big demand, lots of pretenders, and very few 10X coders and thinkers are available. AI Marketers? A surplus because math and physics are hard and art history and social science are somewhat less demanding on today’s thumb typers.
- Tools, lots of tools. Who has time to keep track of every “new” piece of smart software tooling? I gave up as the hyperbole got underway in early 2023. When my team needs to do something specific, they look / hunt for possibilities. Testing is required because smart software often gets things wrong. Some call this “innovation.” I call it evidence of the proliferation of flawed or cute software. One cannot machine titanium with lousy tools.
- Management measurements. Give me a break, Ms. Farmer. Managers are often evidence of the Peter Principle, an accountant, or a lawyer. How can one measure what one does not use, understand, or creates? Those chasing smart software are not making spindles for a wooden staircase. The task of creating smart software that has a shot at producing money is neither art nor science. It is a continuous process of seeing what works, fiddling, and fumbling. You want to measure this? Good luck, although blue chip consultants will gladly create a slide deck to show you the ropes and then churn out a spectacular invoice for professional services.
One question: Is GitLab part of the problem or part of the solution?
Stephen E Arnold, January 9, 2025
AI Outfit Pitches Anti Human Message
January 9, 2025
AI startup Artisan thought it could capture attention by telling companies to get rid of human workers and use its software instead. It was right. Gizmodo reports, “AI Firm’s ‘Stop Hiring Humans’ Billboard Campaign Sparks Outrage.” The firm plastered its provocative messaging across San Francisco. Writer Lucas Ropek reports:
“The company, which is backed by startup accelerator Y-Combinator, sells what it calls ‘AI Employees’ or ‘Artisans.’ What the company actually sells is software designed to assist with customer service and sales workflow. The company appears to have done an internal pow-wow and decided that the most effective way to promote its relatively mundane product was to fund an ad campaign heralding the end of the human age. Writing about the ad campaign, local outlet SFGate notes that the posters—which are strewn all over the city—include plugs like the following:
‘Artisans won’t complain about work-life balance’
‘Artisan’s Zoom cameras will never ‘not be working’ today.’
‘Hire Artisans, not humans.’
‘The era of AI employees is here.'”
The write-up points to an interview with SFGate in which CEO Jaspar Carmichael-Jack states the ad campaign was designed to “draw eyes.” Mission accomplished. (And is it just me, or does that name belong in a pirate movie?) Though Ropek acknowledges his part in drawing those eyes, he also takes this chance to vent about AI and big tech in general. He writes:
“It is Carmichael-Jackson’s admission that his billboards are ‘dystopian’—just like the product he’s selling—that gets to the heart of what is so [messed] up about the whole thing. It’s obvious that Silicon Valley’s code monkeys now embrace a fatalistic bent of history towards the Bladerunner-style hellscape their market imperatives are driving us.”
Like Artisan’s billboards, Ropek pulls no punches. Located in San Francisco, Artisan was launched in 2023. Founders hail from the likes of Stanford, Oxford, Meta, and IBM. Will the firm find a way to make its next outreach even more outrageous?
Cynthia Murrell, January 9, 2025
Ground Hog Day: Smart Enterprise Search
January 7, 2025
I am a dinobaby. I also wrote the Enterprise Search Report, 1st, 2nd, and 3rd editions. I wrote The New Landscape of Search. I wrote some other books. The publishers are long gone, and I am mostly forgotten in the world of information retrieval. Read this post, and you will learn why. Oh, no AI helped me out unless I come up with an art idea. I used Stable Diffusion for the rat, er, sorry, ground hog day creature.
I think it was 2002 when the owner of a publishing company asked me if I thought there was an interest in profiles of companies offering “enterprise search solutions.” I vaguely remember the person, and I will leave it up to you to locate a copy of the 400 page books I wrote about enterprise search.
The set up for the book was simple. I identified the companies which seemed to bid on government contracts for search, companies providing search and retrieval to organizations, and outfits which had contacted me to pitch their enterprise search systems before they were exiting stealth mode. By the time the first edition appeared in 2004, the companies in the ESR were flogging their products.
The ground hog effect is a version of the Yogi Berra “Déjà vu all over again” thing. Enterprise search is just out of reach now and maybe forever.
The enterprise search market imploded. It was there and then it wasn’t. Can you describe the features and functions of these enterprise search systems from the “golden age” of information retrieval:
- Innerprise
- InQuira
- iPhrase
- Lextek Onix
- MondoSearch
- Speed of Mind
- Stratify (formerly Purple Yogi)
The end of enterprise search coincided with large commercial enterprises figuring out that “search” in a complex organization was not one thing. The problem remains today. Lawyers in a Fortune 1000 company want one type of search. Marketers want another “flavor” of search. The accountants want a search that retrieves structured and unstructured data plus images of invoices. Chemists want chemical structure search. Senior managers want absolutely zero search of their personal and privileged data unless it is lawyers dealing with litigation. In short, each unit wants a highly particularized search and each user wants access to his or her data. Access controls are essential, and they are a hassle at a time when the notion of an access control list was like learning to bake bread following a recipe in Egyptian hieroglyphics.
These problems exist today and are complicated by podcasts, video, specialized file types for 3D printing, email, encrypted messaging, unencrypted messaging, and social media. No one has cracked the problem of a senior sales person who changes a PowerPoint deck to close a deal. Where is that particular PowerPoint? Few know and the sales person may have deleted the file changed minutes before the face to face pitch. This means that baloney like “all” the information in an organization is searchable is not just stupid; it is impossible.
The key events were the legal and financial hassles over Fast Search & Transfer. Microsoft bought the company in 2008 and that was the end of a reasonably capable technology platform and — believe it or not — a genuine alternative to Google Web search. A number of enterprise search companies sold out because the cost of keeping the technology current and actually running a high-grade sales and marketing program spelled financial doom. Examples include Exalead and Vivisimo, among others. Others just went out of business: Delphes (remember that one?). The kiss of death for the type of enterprise search emphasized in the ESR was the acquisition of Autonomy by Hewlett Packard. There was a roll up play underway by OpenText which has redefined itself as a smart software company with Fulcrum and BRS Search under its wing.
What replaced enterprise search when the dust settled in 2011? From my point of view it was Shay Banon’s Elastic search and retrieval system. One might argue that Lucid Works (né Lucid Imagination) was a player. That’s okay. I am, however, to go with Elastic because it offered a version as open source and a commercial version with options for on-going engineering support. For the commercial alternatives, I would say that Microsoft became the default provider. I don’t think SharePoint search “worked” very well, but it was available. Google’s Search Appliance appeared and disappeared. There was zero upside for the Google with a product that was “inefficient” at making a big profit for the firm. So, Microsoft it was. For some government agencies, there was Oracle.
Oracle acquired Endeca and focused on that computationally wild system’s ability to power eCommerce sites. Oracle paid about $1 billion for a system which used to be an enterprise search with consulting baked in. One could buy enterprise search from Oracle and get structured query language search, what Oracle called “secure enterprise search,” and may a dollop of Triple Hop and some other search systems the company absorbed before the end of the enterprise search era. IBM talked about search but the last time I drove by IBM Government systems in Gaithersburg, Maryland, it like IBM search, had moved on. Yo, Watson.
Why did I make this dalliance on memory lane the boring introduction to a blog post? The answer is that I read “Are LLMs At Risk Of Going The Way Of Search? Expect A Duopoly.” This is a paywalled article, so you will have to pony up cash or go to a library. Here’s an abstract of the write up:
The evolution of LLMs (Large Language Models) will lead users to prefer one or two dominant models, similar to Google’s dominance in search.
Companies like Google and Meta are well-positioned to dominate generative AI due to their financial resources, massive user bases, and extensive data for training.
Enterprise use cases present a significant opportunity for specialized models.
Therefore, consumer search will become a monopoly or duopoly.
Let’s assume the Forbes analysis is accurate. Here’s what I think will happen:
First, the smart software train will slow and a number of repackagers will use what’s good enough; that is, cheap enough and keeps the client happy. Thus, a “golden age” of smart search will appear with outfits like Google, Meta, Microsoft, and a handful of others operating as utilities. The US government may standardize on Microsoft, but it will be partners who make the system meet the quite particular needs of a government entity.
Second, the trajectory of the “golden age” will end as it did for enterprise search. The costs and shortcomings become known. Years will pass, probably a decade, maybe less, until a “new” approach becomes feasible. The news will diffuse and then a seismic event will occur. For AI, it was the 2023 announcement that Microsoft and OpenAI would change how people used Microsoft products and services. This created the Google catch up and PR push. We are in the midst of this at the start of 2025.
Third, some of the problems associated with enterprise information and an employee’s finding exactly what he or she needs will be solved. However, not “all” of the problems will be solved. Why? The nature of information is that it is a bit like pushing mercury around. The task requires fresh thinking.
To sum up, the problem of search is an excellent illustration of the old Hegelian chestnut of Hegelian thesis, antithesis, and synthesis. This means the problem of search is unlikely to be “solved.” Humans want answers. Some humans want to verify answers which means that the data on the sales person’s laptop must be included. When the detail oriented human learns that the sales person’s data are missing, the end of the “search solution” has begun.
The question “Will one big company dominate?” The answer is, in my opinion, maybe in some use cases. Monopolies seem to be the natural state of social media, online advertising, and certain cloud services. For finding information, I don’t think the smart software will be able to deliver. Examples are likely to include [a] use cases in China and similar countries, [b] big multi-national organizations with information silos, [c] entities involved in two or more classified activities for a government, [d] high risk legal cases, and [e] activities related to innovation, trade secrets, and patents, among others.
The point is that search and retrieval remains an extraordinarily difficult problem to solve in many situations. LLMs contribute some useful functional options, but by themselves, these approaches are unlikely to avoid the reefs which sank the good ships Autonomy and Fast Search & Transfer, and dozens of others competing in the search space.
Maybe Yogi Berra did not say “Déjà vu all over again.” That’s okay. I will say it. Enterprise search is “Déjà vu all over again.”
Stephen E Arnold, January 7, 2025
-
- Subscribe to Beyond Search
Feature archive
News archive
-
