AI Snake Oil Hisses at AI
August 23, 2024
![green-dino_thumb_thumb_thumb_thumb_t[1]_thumb](https://arnoldit.com/wordpress/wp-content/uploads/2024/08/green-dino_thumb_thumb_thumb_thumb_t1_thumb-3.gif "green-dino_thumb_thumb_thumb_thumb_t[1]_thumb")
This essay is the work of a dumb dinobaby. No smart software required.
Enthusiasm for certain types of novel software or gadgets rises and falls. The Microsoft marketing play with OpenAI marked the beginning of the smart software hype derby. Google got the message and flipped into Red Alert mode. Now about 20 months after Microsoft’s announcement about its AI tie up with Sam AI-Man, we have Google’s new combo: AI in a mobile phone. Bam! Job done. Slam dunk.
Thanks, MSFT Copilot. On top of the IPv6 issue? Oh, too bad.
I wonder if the Googlers were thinking along the same logical lines at the authors of “AI Companies Are Pivoting from Creating Gods to Building Products. Good.”
The snake oil? Dripping. Here’s a passage from the article I noted:
AI companies are collectively planning to spend a trillion dollars on hardware and data centers, but there’s been relatively little to show for it so far.
A trillion? That’s a decent number. Sam AI-Man wants more, but the scale is helpful, particularly when most numbers are mere billions in the zoom zoom world of smart software.
The most important item in the write up, in my opinion, is the list of five “challenges.” The article focuses on consumer AI. A couple of these apply to the enterprise sector as well. Let’s look at the five “challenges.” These are and, keep in mind, I a paraphrasing as dinobabies often do:
- Cost. In terms of consumers, one must consider making Hamster Kombat smart. (This is a Telegram dApp.) My team informed me that this little gem has 35 million users, and it is still growing. Imagine the computational cost to infuse each and every Hamster Kombat “game” player with AI goodness. But it’s a game and a distributed one at that, one might say. Someone has to pay for these cycles. And Hamster Kombat is not on the radar of most consumers’ radar. Telegram has about 950 million users, so 35 million users comes from that pool. What are the costs of AI infused games outside of a walled garden. And the hardware? And the optimization engineering? And the fooling around with ad deals? Costs are not a hurdle. Costs might be a Grand Canyon-scale leap into a financial mud bank.
- Reliability. Immature systems and methods, training content issues (real and synthetic), and the fancy math which uses a lot of probability procedures guarantees some interesting outputs.
- Privacy. The consumer or user facing services are immature. Developers want to get something to most work in a good enough manner. Then security may be discussed. But on to the next feature. As a result, I am not sure if anyone has a decent grasp of the security issues which smart software might pose. Look at Microsoft. It’s been around almost half a century, and I learn about new security problems every day. Is smart software different?
- Safety and security. This is a concomitant to privacy. Good luck knowing what the systems do or do not do.
- User interface. I am a dinobaby. The interfaces are pale, low contrast, and change depending on what a user clicks. I like stability. Smart software simply does not comprehend that word.
Good points. My view is that the obstacle to surmount is money. I am not sure that the big outfits anticipated the costs of their sally into the hallucinating world of AI. And what are those costs, pray tell. Here’s are selected items the financial managers at the Big Dogs are pondering along with the wording of their updated LinkedIn profile:
- Litigation. Remarks by some icons of the high technology sector have done little to assuage the feelings of those whose content was used without permission or compensation. Some, some people. A few Big Dogs are paying cash to scrape.
- Power. Yep, electricity, as EV owners know, is not really free.
- Water, Yep, modern machines produce heat if what I learned in physics was actual factual.
- People (until they can be replaced by a machine that does not require health care or engage in signing petitions).
- Data and indexing. Yep, still around and expensive.
- License fees. They are comin’ round the mountain of legal filings.
- Meals, travel and lodging. Leadership will be testifying, probably a lot.
- PR advisors and crisis consultants. See the first bullet, Litigation.
However, slowly but surely some commercial sectors are using smart software. There is an AI law firm. There are dermatologists letting AI determine what to cut, freeze, or ignore. And there are college professors using AI to help them do “original” work and create peer-review fodder.
There was a snake in the Garden of Eden, right?
Stephen E Arnold, August 23, 2024
AI Research: A New and Slippery Cost Center for the Google
August 7, 2024
This essay is the work of a dumb humanoid. No smart software required.
A week or so ago, I read “Scaling Exponents Across Parameterizations and Optimizers.” The write up made crystal clear that Google’s DeepMind can cook up a test, throw bodies at it, and generate a bit of “gray” literature. The objective, in my opinion, was three-fold. [1] The paper makes clear that DeepMind is thinking about its smart software’s weaknesses and wants to figure out what to do about them. And [2] DeepMind wants to keep up the flow of PR – Marketing which says, “We are really the Big Dogs in this stuff. Good luck catching up with the DeepMind deep researchers.” Note: The third item appears after the numbers.
I think the paper reveals a third and unintended consequence. This issue is made more tangible by an entity named 152334H and captured in “Calculating the Cost of a Google DeepMind Paper.” (Oh, 152334 is a deep blue black color if anyone cares.)
That write up presents calculations supporting this assertion:
How to burn US$10,000,000 on an arXiv preprint
The write up included this table presenting the costs to replicate what the xx Googlers and DeepMinders did to produce the ArXiv gray paper:
Notice, please, that the estimate is nearly $13 million. Anyone want to verify the Google results? What am I hearing? Crickets.
The gray paper’s 11 authors had to run the draft by review leadership and a lawyer or two. Once okayed, the document was converted to the arXiv format, and we the findings to improve our understanding of how much work goes into the achievements of the quantumly supreme Google.
Thijs number of $12 million and change brings me to item [3]. The paper illustrates why Google has a tough time controlling its costs. The paper is not “marketing,” because it is R&D. Some of the expense can be shuffled around. But in my book, the research is overhead, but it is not counted like the costs of cubicles for administrative assistants. It is science; it is a cost of doing business. Suck it up, you buttercups, in accounting.
The write up illustrates why Google needs as much money as it can possibly grab. These costs which are not really nice, tidy costs have to be covered. With more than 150,000 people working on projects, the costs of “gray” papers is a trigger for more costs. The compute time has to be paid for. Hello, cloud customers. The “thinking time” has to be paid for because coming up with great research is open ended and may take weeks, months, or years. One could not rush Einstein. One cannot rush Google wizards in the AI realm either.
The point of this blog post is to create a bit of sympathy for the professionals in Google’s accounting department. Those folks have a tough job figuring out how to cut costs. One cannot prevent 11 people from burning through computer time. The costs just hockey stick. Consequently the quantumly supreme professionals involved in Google cost control look for simpler, more comprehensible ways to generate sufficient cash to cover what are essentially “surprise” costs. These tools include magic wand behavior over payments to creators, smart commission tables to compensate advertising partners, and demands for more efficiency from Googlers who are not thinking big thoughts about big AI topics.
Net net: Have some awareness of how tough it is to be quantumly supreme. One has to keep the PR and Marketing messaging on track. One has to notch breakthroughs, insights, and innovations. What about that glue on the pizza thing? Answer: What?
Stephen E Arnold, August 7, 2024
The Truth Behind Cloud Computing Costs
January 25, 2013
Cloud computing allows users ease and accessibility and lower costs, but Datamation analyzes, “What Are The Hidden Costs Of Cloud Computing?” Datamation pulls it information from Symantec’s “Avoiding the Hidden Costs of Cloud 2013” survey, that shows cloud adoption is very high. Rogue deployments came from 77% of the survey-takers, meaning the clouds were not approved by corporate IT. Does anyone else here security breech problems?
Cloud storage may be lighter than air, but it is also costing money. That is not the scary point, however:
“Perhaps even more concerning is the fact that 43 percent of respondents admitted that they have lost data in the cloud. In Elliot’s view, there are a number of reasons for the cloud data loss. For one, the cloud data provider could have lost the data in a failure of some sort. What is more likely, though, is that some form of user human error led to the data loss.
‘The user could have accidentally misplaced the data and literally just could not find it,’ [Dave Elliott, senior manager, Global Cloud Marketing at Symantec] said.”
Cloud users are also not fully using all of the storage they pay for and they also do not have data duplication in place. What can we learn from this? Install security policies for the cloud, reevaluate the costs, and make sure everything is consistent! In truth, does anyone even notice the costs compared to responsibilities being shuffled around and cutting staff?
Whitney Grace, January 25, 2013
Sponsored by ArnoldIT.com, developer of Beyond Search
Amazon Cloud Cost Comparison Revealed
December 28, 2012
Deep Value, purveyor of respected stock-trading algorithms, is dissatisfied with its Amazon Cloud Services investment. The company’s blog declares, “EC2 is 380% More Expensive than Internal Cluster.” Deep Value was using Amazon’s EC2 cluster with Hadoop to run simulations, but the bill kept increasing month by month. Their managing director of technology, Paul Haefele, says he performed some “back of the envelope” calculations that suggested there was a better way. He writes:
“Tiger Direct will sell you a Seagate 3 terabyte drive for $154. For the same storage on S3 for 2 years, I would pay (1,000 * 0.125 + 2,000 * 0.11) * 12 mths * 2 yrs1 = $8,232 at the standard rates. Buying our own drive was 2% of the cost of using EC2, so this certainly seemed worth investigating.”
Haefele’s team deployed a Hadoop cluster with the data-center company Telx, whose rates they found reasonable. They invested in 20 Linux servers running the open-source CentOS and a couple of switches. They factored in hosting costs, then ran some simulations. See the article for the technical details; Haefele summarizes his conclusions:
“If we compared just on what we are getting in terms of compute and storage, our cluster is costing us $12,700 per month versus $48,564 (33,599+15,965) for EC2.
“EC2 is thus costing us over 3.8 time more per month.
“Whatever way we slice this, either by storage cost or by compute, it seems clear that using your own data center rather than EC2 makes sense for us. For one-off peaks EC2 makes sense, but given the ongoing nature of our simulated analysis, moving to our own datacenter is a very clear winner.”
So, sometimes DIY is worth the effort, as this extreme example shows. Thanks to Deep Value for sharing this important information. I wonder—how many businesses will take advantage of it?
Cynthia Murrell, December 28, 2012
Sponsored by ArnoldIT.com, developer of Augmentext
Quote to Note: Real Consultant Illuminates HP Autonomy
November 22, 2012
Heck, let’s put the 451 Group quote in granite.
I am getting ready to head from England, the land my ancestors fled in 1596, to return to Harrod’s Creek, Kentucky, the culinary center for down home Thanksgiving cooking. (We do squirrel stew. How about you? Now I am sitting in the Dublin airport relishing comments allegedly made by a “real” expert, a 451 Group consultant no less. My source is “HP-Autonomy Fraud Allegations Fallout: The Winners and Losers.” Please, read the original article. It is priceless.
Here is what has me seeing those cartoon stars and exclamation points popular in the funny papers:
The fraud scandal “probably will” affect Autonomy’s fortunes “and I think that’s a shame…Autonomy always had very good technology, but they weren’t able to leverage it very effectively. In HP’s hands, it could be leveraged much better. On the other hand, Autonomy was “acting as a holding company to a large extent, anyway,” following many acquisitions. Only its core IDOL product is quite closely linked to the Autonomy name, “There are some good pieces there and still some strong brands.” (Attributed to a real consultant doing business as Alan Pelz-Sharp.)
Yowza. Consider these observations I jotted down.
- The word “probably”. Better safe than confident in one’s facts.
- I absolutely love “shame”. Yep, investment bankers are really sensitive to shame, particularly a year after a deal closes and the fees have been paid.
- The assertion “they [Autonomy] were not able to leverage it.” Stunning. I thought Autonomy had been in business for 13 or 14 years and had generated north of $700 million, bought a number of companies, and got its logo on a football team and an F1 car. To my albeit limited knowledge, no other search vendor has come close to Autonomy’s top line revenue. $700 million is a healthy figure and somewhere along the line taxes had to be paid, financial reviews conducted, and invoices sent.
- Autonomy was a holding company. Okay, this is not the analogy I would choose but let’s assume Autonomy was a holding company. Perhaps the 451 Group’s expert knows line of business revenues and pre tax profits by each of Autonomy’s operating units. Where are the breakdowns?
- Only IDOL is linked to Autonomy. My obviously misguided notion was that Autonomy bought a company and then slapped the IDOL brand on the company’s products and services. Then integration tagged along but Autonomy management worked hard to make IDOL the brand for many of its services as Autonomy managed the heck out of staff costs and embarked on upselling. Guess I was ill informed.
- “There are some good pieces” begs some detail. Perhaps a good piece is Virage or the virtual reality Aurasma technology which I covered in one of my Information Today columns?
- Again that categorical “only” troubles me but I will soldier on.
- Yes, I presume fraud will have some impact.
- I find the Latinate predicates stimulating as well.
Some color is in order.
451 Group (the “number” references either a science fiction novel or the flash point of paper, which is going out of fashion among some digerati) is a consulting and professional services firm with expertise in “cloudscape,” “multitenant data centers, and a half dozen other specialties. Disclaimer. I used to work for Booz, Allen & Hamilton and now I write “reports” for IDC, another outfit in the same line of work at 451. This means that one has to determine which consulting firm is on the beam. I will let the reader decide, but my bet is on folks who have done time at one of the pantheon firms, not the firms lower in the rental brains’ pecking order.
I have commented on Autonomy is a number of published reports. These range from the now hoary Enterprise Search Report which first appeared in 2003 or 2004 to my 2011 study “The New Landscape of Enterprise Search.” Free download here. I had to sit through various briefings, do some independent verification and validation work, and interact with some of the folks who had a little first hand experience with things Autonomy. I have to check my files, but I think I did a tiny job for Autonomy, but it is tough for me to remember after 40 years of blundering around in the online, search, and content processing sectors. Unlike the 451 Group’s experts, I am not qualified in the cloudscape thing or the multi whozit. I am so so in search and okay in a couple of other disciplines. I eat squirrels too.
My official position is that folks who license enterprise search systems like Autonomy or Endeca usually wind up with some interesting challenges and some hefty bills. When one taps Autonomy-like system or even a free, open source search solution, the likelihood of financial excitement six months into a search project is quite high. When signing on for an enterprise search system, arrogance and ignorance on the part of the customer can equal the spirit of a gaggle of Googlers. One difference: The Googlers are usually okay with resources, money, code and math. Procurement teams, in my experience, assume their intellectual firepower would decimate other mortals. Wrong. But optimism can be a positive attribute. Just not in big time search solutions.
Enter the marketers, lawyers, and MBAs.
Now a covey of these professionals can draw fascinating conclusions about enterprise search. My problem with the write up and the quote is that the facts of the matter are stark. HP paid billions for a search vendor. HP is in a bit of a financial muddle. HP wants to blame everyone except itself. The losers omitted from the write up are the ones which matter: HP employees, HP management, and the HP Board of Directors.
Maybe the accounting was erroneous. Maybe the auditors stayed up late and drifted through the various audits. Maybe the lawyers, MBAs, and auditors were more focused on their billable hours and commissions. Maybe Autonomy explained how search and content processing deals worked and assumed that the HP team “understood” the upsides and downsides of search.
Somewhere along the line “real” consultants and the legions of people grabbing a piece of the billion dollar deal may be revealing their own intellectual shortcomings. Search is tough. The price HP paid for Autonomy is what it is. A year after buying a property, HP wants its money back. That’s the charming world of business today.
With “real” consultants and big companies looking for a silver bullet to complete their imaginary magic wands, bad deals are the rule. What makes Autonomy interesting is the amount paid by a company which is fraught with management, financial, and technical challenges. I used to keep a list of missteps. But after fumbling the Alta Vista search system opportunity, I lost interest. HP and the “real” consultants want to explain away a mistake without pointing out who triggered the problem. Arrogance, superficial analysis, and lack of search expertise caught fire when big money was in play. Combustible mixture? Yes, indeed.
Stephen E Arnold, November 22, 2012
Science Information Breaking Free From Traditional Publishing
October 21, 2012
A couple of Cambridge entrepreneurs are putting forth another effort to break science information free from traditional publishing’s grip. I recently read the article “A Plan to Open Up Science Journals” on The Boston Globe, which informs us of the project that is attempting to bring the iTunes single-purchase sales approach to often costly scientific journal research. The project is dubbed ReadCube Access and would break the trend of purchasing yearlong subscriptions, called site licenses, by instead offering individual articles for sale.
The article tells us more about the implementation of the product:
“So far, the two entrepreneurs, who are founders of a Cambridge company called Labtiva, have sold the ReadCube Access idea to the industry giant Nature Publishing Group and to the University of Utah’s library system, which started implementing it this fall.
Researchers at the University of Utah can get access to individual journal articles in one of Nature’s 80 or so subscription-based publications, many of which Utah cannot afford to buy.”
This effort to change the way scientific research is published could help libraries lower costs over time. Researchers would also benefit from this method, because ultimately their tedious searchers would result in the source document appearing, not a subscription or purchase form.
Andrea Hayden, October 21, 2012
Sponsored by ArnoldIT.com, developer of Augmentext
The Cost Angle in the Social Media Monitoring Craze
March 20, 2012
Search Engine Journal offers good advice in the very detailed article “Think Before You Pay Too Much for Social Media Monitoring.” Writer Jen Carroll explains:
“Paid tools run the gamut, from blessedly affordable (for solopreneurs and small businesses) to outrageously expensive (only the big guys can even hope to afford them)
“Just like Web analytics tools, social media reporting systems can provide a wealth of data. But, is all that data necessary? How much of it is even useful? For most small and even medium-sized organizations, I would argue that only a small slice of what most reporting tools spit out is actionable.”
Carroll follows up with specific advice under several headings: Track Only the Social Media Metrics That Matter; Try Free Social Media Monitoring Options First; and Carefully Evaluate Paid Social Media Monitoring Tools Before Purchase.
Sample advice: figure out what you need first, so you don’t pay for what you won’t use. Also this tip: pay attention to ease of use; you’ll use the tools more often if you actually like using them.
See the article for more words of wisdom on getting the best value out of social media monitoring.
Stephen E. Arnold, March 20, 2012
Sponsored by Pandia.com
PLM Eases Cost of Change Complexity
January 6, 2012
No one likes change and for manufacturers it can be costly. So how do companies manage and budget for change processes? The article “PLM, ECO and Cost of Change” analyzes that question and how PLM systems support them.
For PLM implementation, it is important to know “how you are able to automate cost of change calculation and embed it in the overall change process.” Changes are classified into four categories which are cost reduction, product maturity, product development and other and the change cost calculation will be different for each category.
“If you estimating change that marked to save cost or time, you absolutely need to calculate the cost. However, if you making a change that related to product maturity, you probably can skip some cost of change calculation. Taking right assumption can significantly improve the speed of change processes, which is an essential part of every manufacturing organization.”
Change cost analysis is complicated. PLM systems certainly help, but companies like Inforbix simplify the process. Inforbix focuses on product data and they find and infer relationships between disparate sources of structured and unstructured product data. So do not be afraid of change. Sometimes change is good.
Jennifer Wensink, January 6, 2012
SharePoint and Its Sometimes Interesting Costs
November 3, 2011
SharePoint is touted as the ultimate solution to content management and collaboration for enterprises. Microsoft, however, never discusses the costs associated with their software, except for how it’s cost effective and overtime will save your business money. But is that true? Redmondmag.com posted an enlightening article about the hidden costs involved in a SharePoint project, “Study: SharePoint Costs High Due to Inadequate Skills.”
A study conducted by the Azaleos Corp. discovered that the average cost to run SharePoint per user is $46/month. Using Microsoft Exchange proved to be cheaper at $15-15/month per user. SharePoint users also cited downtime as the most common problem.
“The downtime mostly stemmed from hardware errors or mistakes made by IT team members. Those problems caused average monthly management costs for SharePoint to double to around $90 per user per month. Almost half (43 percent) of study respondents pointed to “a lack of administrator skills, training, and knowledge as an inhibitor to efficiently leveraging SharePoint.”
SharePoint is still a young piece of software with a manifest destiny for its future. Its problems are many, but there are a lot of third party solutions to resolve them. At the end of the article, Azaleos Corp. advertises it’s AzaleosX app to help increase uptime.
We believe that you may want to take a close look at the cost effective search and content processing solution from SurfRay. Contain costs and improve user satisfaction with one snap in for SharePoint.
Whitney Grace, November 3, 2011
SurfRay
Google Wants More Cash – Enter Google Analytics Premium
October 13, 2011
So basically for a fee, Google will provide a stronger server to run your custom reports, a dedicated account manager and a 24/7 support and service level agreement (SLA). The one thing Google didn’t note in their announcement post was the cost for the premium analytics offering. I have received tips from several sources noting that the price for the premium analytics service is A WHOPPING $150,000/year.
-
- Subscribe to Beyond Search
Feature archive
News archive
-
