Smart Software Project Road Blocks: An Up-to-the-Minute Report
October 1, 2024
![green-dino_thumb_thumb_thumb_thumb_t[2]_thumb_thumb](https://arnoldit.com/wordpress/wp-content/uploads/2024/09/green-dino_thumb_thumb_thumb_thumb_t2_thumb_thumb-2.gif "green-dino_thumb_thumb_thumb_thumb_t[2]_thumb_thumb")
This essay is the work of a dumb dinobaby. No smart software required.
I worked through a 22-page report by SQREAM, a next-gen data services outfit with GPUs. (You can learn more about the company at this buzzword dense link.) The title of the report is:
The report is a marketing document, but it contains some thought provoking content. The “report” was “administered online by Global Surveyz [sic] Research, an independent global research firm.” The explanation of the methodology was brief, but I don’t want to drag anyone through the basics of Statistics 101. As I recall, few cared and were often good customers for my class notes.
Here are three highlights:
- Smart software and services cause sticker shock.
- Cloud spending by the survey sample is going up.
- And the killer statement: 98 percent of the machine learning projects fail.
Let’s take a closer look at the astounding assertion about the 98 percent failure rate.
The stage is set in the section “Top Challenges Pertaining to Machine Learning / Data Analytics.” The report says:
It is therefore no surprise that companies consider the high costs involved in ML experimentation to be the primary disadvantage of ML/data analytics today (41%), followed by the unsatisfactory speed of this process (32%), too much time required by teams (14%) and poor data quality (13%).
The conclusion the authors of the report draw is that companies should hire SQREAM. That’s okay, no surprise because SQREAM ginned up the study and hired a firm to create an objective report, of course.
So money is the Number One issue.
Why do machine learning projects fail? We know the answer: Resources or money. The write up presents as fact:
The top contributing factor to ML project failures in 2023 was insufficient budget (29%), which is consistent with previous findings – including the fact that “budget” is the top challenge in handling and analyzing data at scale, that more than two-thirds of companies experience “bill shock” around their data analytics processes at least quarterly if not more frequently, that that the total cost of analytics is the aspect companies are most dissatisfied with when it comes to their data stack (Figure 4), and that companies consider the high costs involved in ML experimentation to be the primary disadvantage of ML/data analytics today.
I appreciated the inclusion of the costs of data “transformation.” Glib smart software wizards push aside the hassle of normalizing data so the “real” work can get done. Unfortunately, the costs of fixing up source data are often another cause of “sticker shock.” The report says:
Data is typically inaccessible and not ‘workable’ unless it goes through a certain level of transformation. In fact, since different departments within an organization have different needs, it is not uncommon for the same data to be prepared in various ways. Data preparation pipelines are therefore the foundation of data analytics and ML….
In the final pages of the report a number of graphs appear. Here’s one that stopped me in my tracks:
The sample contained 62 percent user of Amazon Web Services. Number 2 was users of Google Cloud at 23 percent. And in third place, quite surprisingly, was Microsoft Azure at 14 percent, tied with Oracle. A question which occurred to me is: “Perhaps the focus on sticker shock is a reflection of Amazon’s pricing, not just people and overhead functions?”
I will have to wait until more data becomes available to me to determine if the AWS skew and the report findings are normal or outliers.
Stephen E Arnold, October 1, 2024
Salesforce: AI Dreams
September 30, 2024
This essay is the work of a dumb dinobaby. No smart software required.
Big Tech companies are heavily investing in AI technology, including Salesforce. Salesforce CEO Marc Benioff delivered a keynote about his company’s future and the end of an era as reported by Constellation Research: “Salesforce Dreamforce 2024: Takeaways On Agentic AI, Platform, End Of Copilot Era.” Benioff described the copilot era as “hit or miss” and he wants to focus on agentic AI powered by Salesforce.
Constellation Research analyst Doug Henschen said that Benioff made compelling case for Salesforce and Data Cloud being the platform that companies will use to build their AI agents. Salesforce already has metadata, data, app business logic knowledge, and more already programmed in it. While Dream Cloud has data integrated from third-party data clouds and ingested from external apps. Combining these components into one platform without DIY is a very appealing product.
Benioff and his team revamped Salesforce to be less a series of clouds that run independently and more of a bunch of clouds that work together in a native system. It means Salesforce will scale Agentforce across Marketing, Commerce, Sales, Revenue and Service Clouds as well as Tableau.
The new AI Salesforce wants to delete DIY says Benioff:
“‘ DIY means I’m just putting it all together on my own. But I don’t think you can DIY this. You want a single, professionally managed, secure, reliable, available platform. You want the ability to deploy this Agentforce capability across all of these people that are so important for your company. We all have struggled in the last two years with this vision of copilots and LLMs. Why are we doing that? We can move from chatbots to copilots to this new Agentforce world, and it’s going to know your business, plan, reason and take action on your behalf.
It’s about the Salesforce platform, and it’s about our core mantra at Salesforce, which is, you don’t want to DIY it. This is why we started this company.’”
Benioff has big plans for Salesforce and based off this Dreamforce keynote it will succeed. However, AI is still experimental. AI is smart but a human is still easier to work with. Salesforce should consider teaming AI with real people for the ultimate solution.
Whitney Grace, September 30, 2024
AI Maybe Should Not Be Accurate, Correct, or Reliable?
September 26, 2024
This essay is the work of a dumb dinobaby. No smart software required.
Okay, AI does not hallucinate. “AI” — whatever that means — does output incorrect, false, made up, and possibly problematic answers. The buzzword “hallucinate” was cooked up by experts in artificial intelligence who do whatever they can to avoid talking about probabilities, human biases migrated into algorithms, and fiddling with the knobs and dials in the computational wonderland of an AI system like Google’s, OpenAI’s, et al. Even the book Why Machines Learn: The Elegant Math Behind Modern AI ends up tangled in math and jargon which may befuddle readers who stopped taking math after high school algebra or who has never thought about Orthogonal matrices.
The Next Web’s “AI Doesn’t Hallucinate — Why Attributing Human Traits to Tech Is Users’ Biggest Pitfall” is an interesting write up. On one hand, it probably captures the attitude of those who just love that AI goodness by blaming humans for anthropomorphizing smart software. On the other hand, the AI systems with which I have interacted output content that is wrong or wonky. I admit that I ask the systems to which I have access for information on topics about which I have some knowledge. Keep in mind that I am an 80 year old dinobaby, and I view “knowledge” as something that comes from bright people working of projects, reading relevant books and articles, and conference presentations or meeting with subjects far from the best exercise leggings or how to get a Web page to the top of a Google results list.
Let’s look at two of the points in the article which caught my attention.
First, consider this passage which is a quote from and AI expert:
“Luckily, it’s not a very widespread problem. It only happens between 2% to maybe 10% of the time at the high end. But still, it can be very dangerous in a business environment. Imagine asking an AI system to diagnose a patient or land an aeroplane,” says Amr Awadallah, an AI expert who’s set to give a talk at VDS2024 on How Gen-AI is Transforming Business & Avoiding the Pitfalls.
Where does the 2 percent to 10 percent number come from? What methods were used to determine that content was off the mark? What was the sample size? Has bad output been tracked longitudinally for the tested systems? Ah, so many questions and zero answers. My take is that the jargon “hallucination” is coming back to bite AI experts on the ankle.
Second, what’s the fix? Not surprisingly, the way out of the problem is to rename “hallucination” to “confabulation”. That’s helpful. Here’s the passage I circled:
“It’s really attributing more to the AI than it is. It’s not thinking in the same way we’re thinking. All it’s doing is trying to predict what the next word should be given all the previous words that have been said,” Awadallah explains. If he had to give this occurrence a name, he would call it a ‘confabulation.’ Confabulations are essentially the addition of words or sentences that fill in the blanks in a way that makes the information look credible, even if it’s incorrect. “[AI models are] highly incentivized to answer any question. It doesn’t want to tell you, ‘I don’t know’,” says Awadallah.
Third, let’s not forget that the problem rests with the users, the personifies, the people who own French bulldogs and talk to them as though they were the favorite in a large family. Here’s the passage:
The danger here is that while some confabulations are easy to detect because they border on the absurd, most of the time an AI will present information that is very believable. And the more we begin to rely on AI to help us speed up productivity, the more we may take their seemingly believable responses at face value. This means companies need to be vigilant about including human oversight for every task an AI completes, dedicating more and not less time and resources.
The ending of the article is a remarkable statement; to wit:
As we edge closer and closer to eliminating AI confabulations, an interesting question to consider is, do we actually want AI to be factual and correct 100% of the time? Could limiting their responses also limit our ability to use them for creative tasks?
Let me answer the question: Yes, outputs should be presented and possibly scored; for example, 90 percent probable that the information is verifiable. Maybe emojis will work? Wow.
Stephen E Arnold, September 26, 2024
AI Automation Has a Benefit … for Some
September 26, 2024
Humanity’s progress runs parallel to advancing technology. As technology advances, aspects of human society and culture are rendered obsolete and it is replaced with new things. Job automation is a huge part of this; past example are the Industrial Revolution and the implementation of computers. AI algorithms are set to make another part of the labor force defunct, but the BBC claims that might be beneficial to workers: “Klarna: AI Lets Us Cut Thousands Of Jobs-But Pay More.”
Klarna is a fintech company that provides online financial services and is described as a “buy now, pay later” company. Klarna plans to use AI to automate the majority of its workforce. The company’s leaders already canned 1200 employees and they plan to fire another 2000 as AI marketing and customer service is implemented. That leaves Klarna with a grand total of 1800 employees who will be paid more.
Klarna’s CEO Sebastian Siematkowski is putting a positive spin on cutting jobs by saying the remaining employees will receive larger salaries. While Siematkowski sees the benefits of AI, he does warn about AI’s downside and advises the government to do something. He said:
“ ‘I think politicians already today should consider whether there are other alternatives of how they could support people that may be effective,’ he told the Today programme, on BBC Radio 4.
He said it was “too simplistic” to simply say new jobs would be created in the future.
‘I mean, maybe you can become an influencer, but it’s hard to do so if you are 55-years-old,’ he said.”
The International Monetary Fund (IMF) predicts that 40% of all jobs will worsen in “overall equality” due to AI. As Klarna reduces its staff, the company will enter what is called “natural attrition” aka a hiring freeze. The remaining workforce will have bigger workloads. Siematkowski claims AI will eventually reduce those workloads.
Will that really happen? Maybe?
Will the remaining workers receive a pay raise or will that money go straight to the leaders’ pockets? Probably.
Whitney Grace, September 26, 2024
Amazon Has a Better Idea about Catching Up with Other AI Outfits
September 25, 2024
AWS Program to Bolster 80 AI Startups from Around the World
Can boosting a roster of little-known startups help AWS catch up with Google’s and Microsoft’s AI successes? Amazon must hope so. It just tapped 80 companies from around the world to receive substantial support in its AWS Global Generative AI Accelerator program. Each firm will receive up to $1 million in AWS credits, expert mentorship, and a slot at the AWS re:Invent conference in December.
India’s CXOtoday is particularly proud of the seven recipients from that country. It boasts, “AWS Selects Seven Generative AI Startups from India for Global AWS Generative AI Accelerator.” We learn:
“The selected Indian startups— Convrse, House of Models, Neural Garage, Orbo.ai, Phot.ai, Unscript AI, and Zocket, are among the 80 companies selected by AWS worldwide for their innovative use of AI and their global growth ambitions. The Indian cohort also represents the highest number of startups selected from a country in the Asia-Pacific region for the AWS Global Generative AI Accelerator program.”
The post offers this stat as evidence India is now an AI hotspot. It also supplies some more details about the Amazon program:
“Selected startups will gain access to AWS compute, storage, and database technologies, as well as AWS Trainium and AWS Inferentia2, energy-efficient AI chips that offer high performance at the lowest cost. The credits can also be used on Amazon SageMaker, a fully managed service that helps companies build and train their own foundation models (FMs), as well as to access models and tools to easily and securely build generative AI applications through Amazon Bedrock. The 10-week program matches participants with both business and technical mentors based on their industry, and chosen startups will receive up to US$1 million each in AWS credits to help them build, train, test, and launch their generative AI solutions. Participants will also have access to technology and technical sessions from program presenting partner NVIDIA.”
See the write-up to learn more about each of the Indian startups selected, or check out the full roster here.
The question is, “Will this help Amazon which is struggling to make Facebook, Google, and Microsoft look like the leaders in the AI derby?”
Cynthia Murrell, September 25, 2024
Open Source Dox Chaos: An Opportunity for AI
September 24, 2024
It is a problem as old as the concept of open source itself. ZDNet laments, “Linux and Open-Source Documentation Is a Mess: Here’s the Solution.” We won’t leave you in suspense. Writer Steven Vaughan-Nichols’ solution is the obvious one—pay people to write and organize good documentation. Less obvious is who will foot the bill. Generous donors? Governments? Corporations with their own agendas? That question is left unanswered.
But there is not doubt. Open-source documentation, when it exists at all, is almost universally bad. Vaughan-Nichols recounts:
“When I was a wet-behind-the-ears Unix user and programmer, the go-to response to any tech question was RTFM, which stands for ‘Read the F… Fine Manual.’ Unfortunately, this hasn’t changed for the Linux and open-source software generations. It’s high time we addressed this issue and brought about positive change. The manuals and almost all the documentation are often outdated, sometimes nearly impossible to read, and sometimes, they don’t even exist.”
Not only are the manuals that have been cobbled together outdated and hard to read, they are often so disorganized it is hard to find what one is looking for. Even when it is there. Somewhere. The post emphasizes:
“It doesn’t help any that kernel documentation consists of ‘thousands of individual documents’ written in isolation rather than a coherent body of documentation. While efforts have been made to organize documents into books for specific readers, the overall documentation still lacks a unified structure. Steve Rostedt, a Google software engineer and Linux kernel developer, would agree. At last year’s Linux Plumbers conference, he said, ‘when he runs into bugs, he can’t find documents describing how things work.’ If someone as senior as Rostedt has trouble, how much luck do you think a novice programmer will have trying to find an answer to a difficult question?”
This problem is no secret in the open-source community. Many feel so strongly about it they spend hours of unpaid time working to address it. Until they just cannot take it anymore. It is easy to get burned out when one is barely making a dent and no one appreciates the effort. At least, not enough to pay for it.
Here at Beyond Search we have a question: Why can’t Microsoft’s vaunted Copilot tackle this information problem? Maybe Copilot cannot do the job?
Cynthia Murrell, September 24, 2024
Microsoft Explains Who Is at Fault If Copilot Smart Software Does Dumb Things
September 23, 2024
This essay is the work of a dumb dinobaby. No smart software required.
Those Windows Central experts have delivered a Dusie of a write up. “Microsoft Says OpenAI’s ChatGPT Isn’t Better than Copilot; You Just Aren’t Using It Right, But Copilot Academy Is Here to Help” explains:
Avid AI users often boast about ChatGPT’s advanced user experience and capabilities compared to Microsoft’s Copilot AI offering, although both chatbots are based on OpenAI’s technology. Earlier this year, a report disclosed that the top complaint about Copilot AI at Microsoft is that “it doesn’t seem to work as well as ChatGPT.”
I think I understand. Microsoft uses OpenAI, other smart software, and home brew code to deliver Copilot in apps, the browser, and Azure services. However, users have reported that Copilot doesn’t work as well as ChatGPT. That’s interesting. A hallucinating capable software processed by the Microsoft engineering legions is allegedly inferior to Copilot.
Enthusiastic young car owners replace individual parts. But the old car remains an old, rusty vehicle. Thanks, MSFT Copilot. Good enough. No, I don’t want to attend a class to learn how to use you.
Who is responsible? The answer certainly surprised me. Here’s what the Windows Central wizards offer:
A Microsoft employee indicated that the quality of Copilot’s response depends on how you present your prompt or query. At the time, the tech giant leveraged curated videos to help users improve their prompt engineering skills. And now, Microsoft is scaling things a notch higher with Copilot Academy. As you might have guessed, Copilot Academy is a program designed to help businesses learn the best practices when interacting and leveraging the tool’s capabilities.
I think this means that the user is at fault, not Microsoft’s refactored version of OpenAI’s smart software. The fix is for the user to learn how to write prompts. Microsoft is not responsible. But OpenAI’s implementation of ChatGPT is perceived as better. Furthermore, training to use ChatGPT is left to third parties. I hope I am close to the pin on this summary. OpenAI just puts Strawberries in front of hungry users and let’s them gobble up ChatGPT output. Microsoft fixes up ChatGPT and users are allegedly not happy. Therefore, Microsoft puts the burden on the user to learn how to interact with the Microsoft version of ChatGPT.
I thought smart software was intended to make work easier and more efficient. Why do I have to go to school to learn Copilot when I can just pound text or a chunk of data into ChatGPT, click a button, and get an output? Not even a Palantir boot camp will lure me to the service. Sorry, pal.
My hypothesis is that Microsoft is a couple of steps away from creating something designed for regular users. In its effort to “improve” ChatGPT, the experience of using Copilot makes the user’s life more miserable. I think Microsoft’s own engineering practices act like a struck brake on an old Lada. The vehicle has problems, so installing a new master cylinder does not improve the automobile.
Crazy thinking: That’s what the write up suggests to me.
Stephen E Arnold, September 23, 2024
DAIS: A New Attempt to Make AI Play Nicely with Humans
September 20, 2024
This essay is the work of a dumb dinobaby. No smart software required.
How about a decentralized artificial intelligence “association”? One has been set up by Michael Casey, the former chief content officer at Coindesk. (Coindesk reports about the bright, sunny world of crypto currency and related topics.) I learned about this society in — you guessed it — Coindesk’s online information service called Coindesk. The article “Decentralized AI Society Launched to Fight Tech Giants Who ‘Own the Regulators’” is interesting. I like the idea that “tech giants” own the regulators. This is an observation which Apple and Google might not agree. Both “tech giants” have been facing some unfavorable regulatory decisions. If these regulators are “owned,” I think the “tech giants” need to exercise their leadership skills to make the annoying regulators go away. One resigned in the EU this week, but as Shakespeare said of lawyers, let’s drown them. So far the “tech giants” have been bumbling along, growing bigger as a result of feasting on data and amplifying allegedly monopolistic behaviors which just seem to pop up, rules or no rules.
Two experts look at what emerged from a Petri dish of technological goodies. Quite a surprise I assume. Thanks, MSFT Copilot. Good enough.
The write up reports:
Industry leaders have launched a non-profit organization called the Decentralized AI Society (DAIS), dedicated to tackling the probability of the monopolization of the artificial intelligence (AI) industry.
What is the DAIS outfit setting out to do? Here’s what Coindesk reports and this is a quote of the bullets from the write up:
Bringing capital to the decentralized AI world in what has already become an arms race for resources like graphical processing units (GPUs) and the data centers that compute together.
Shaping policy to craft AI regulations.
Education and promotion of decentralized AI.
Engineering to create new algorithms for learning models in a distributed way.
These are interesting targets. I want to point out that “decentralization” is the opposite of what the “tech giants” have already put in place; that is, concentration of money, talent, and infrastructure. Even old dogs like Oracle are now hopping on the centralized bandwagon. Even newcomers want to get as many cattle into the killing chute before the glamor of AI begins to lose some sparkles.
Several observations:
- DAIS has some crypto roots. These may become positive or negative. Right now regulators are interested in crypto as are other enforcement entities
- One of the Arnold Laws of Online is that centralization, consolidation, and concentration are emergent behaviors for online products and services. Countering this “law” and its “emergent” functionality is going to take more than conferences, a Web site, and some “logical” ideas which any “rational” person would heartily endorse. But emergent is tough to stop based on my experience.
- Singapore has become a hot spot for certain financial and technical activities. The problem is that nation-states may not want to be inhibited in their AI ambitions. Some may find the notion of “education” a problem as well because curricula must conform to pre-defined frameworks. Distributed is not a pre-defined anything; it is the opposite of controlled and, therefore, likely to be a bit of a problem.
Net net: Interesting idea. But Amazon, Google, Facebook, Microsoft, and some other outfits may want to talk about “distributed” but really mean the technological notion is okay, but we want as much of the money as we can get.
Stephen E Arnold, September 20, 2024
YouTube Is Bringing More AI To Its Platform
September 20, 2024
AI-generated videos have already swarmed on YouTube. These videos range from fake Disney movie trailers to inappropriate content that missed being flagged. YouTube creators are already upset that their videos are being overlooked by the algorithm, but some are being hired for an AI project. Digital Trends explains more: “More AI May Be Coming To YouTube In A Big Way.”
Gemini AI is currently in beta testing across YouTube. Gemini AI is described as a tool for YouTubers to brainstorm video ideas, including titles, topics, and thumbnails. Only a select few YouTubers are testing Gemini AI and will share their feedback. The AI tool will eventually be located underneath the platform’s analytic menu, under the research tab. The tool could actually be helpful:
“This marks Google’s second foray into including AI assistance in YouTube users’ creative processes. In May, the company launched a content inspiration tool on YouTube Studio that provides tips and suggestions for future clip topics based on viewer trends. For most any given topic, the AI will highlight related videos you’ve already published, provide tips on themes to use, and generate a script outline for you to follow.”
The YouTubers are experimenting with both Gemini AI and the content inspiration tool. They’re doing A/B testing and their experiences will shape how AI is used on the video platform. YouTube does acknowledge that AI is a transformative creative tool, but viewers want to know if what they’re watching is real or fake. Is anyone imagining a AI warning or rating system?
Whitney Grace, September 20, 2024
Happy AI News: Job Losses? Nope, Not a Thing
September 19, 2024
This essay is the work of a dumb humanoid. No smart software required.
I read “AI May Not Steal Many Jobs after All. It May Just Make Workers More Efficient.” Immediately two points jumped out at me. The AP (the publisher of the “real” news story is hedging with the weasel word “may” and the hedgy phrase “after all.” Why is this important? The “real” news industry is interested in smart software to reduce costs and generate more “real” news more quickly. The days with “real” reporters disappearing for hours to confirm with a source are often associated with fiddling around. The costs of doing anything without a gusher of money pumping 24×7 are daunting. The word “efficient” sits in the headline as a digital harridan stakeholder. Who wants that?
The manager of a global news operation reports that under his watch, he has achieved peak efficiency. Thanks, MSFT Copilot. Will this work for production software development? Good enough is the new benchmark, right?
The story itself strikes me as a bit of content marketing which says, “Hey, everyone can use AI to become more efficient.” The subtext is, “Hey, don’t worry. No software robot or agentic thingy will reduce staff. Probably.
The AP is a litigious outfit even though I worked at a newspaper which “participated” in the business process of the entity. Here’s one sentence from the “real” news write up:
Instead, the technology might turn out to be more like breakthroughs of the past — the steam engine, electricity, the internet: That is, eliminate some jobs while creating others. And probably making workers more productive in general, to the eventual benefit of themselves, their employers and the economy.
Yep, just like the steam engine and the Internet.
When technologies emerge, most go away or become componentized or dematerialized. When one of those hot technologies fail to produce revenues, quite predictable outcomes result. Executives get fired. VC firms do fancy dancing. IRS professionals squint at tax returns.
So far AI has been a “big guys win sort of because they have bundles of cash” and “little outfits lose control of their costs”. Here’s my take:
- Human-generated news is expensive and if smart software can do a good enough job, that software will be deployed. The test will be real time. If the software fails, the company may sell itself, pivot, or run a garage sale.
- When “good enough” is the benchmark, staff will be replaced with smart software. Some of the whiz kids in AI like the buzzword “agentic.” Okay, agentic systems will replace humans with good enough smart software. That will happen. Excellence is not the goal. Money saving is.
- Over time, the ideas of the current transformer-based AI systems will be enriched by other numerical procedures and maybe— just maybe — some novel methods will provide “smart software” with more capabilities. Right now, most smart software just finds a path through already-known information. No output is new, just close to what the system’s math concludes is on point. Right now, the next generation of smart software seems to be in the future. How far? It’s anyone’s guess.
My hunch is that Amazon Audible will suggest that humans will not lose their jobs. However, the company is allegedly going to replace human voices with “audibles” generated by smart software. (For more about this displacement of humans, check out the Bloomberg story.)
Net net: The “real” news story prepares the field for planting writing software in an organization. It says, “Customer will benefit and produce more jobs.” Great assertions. I think AI will be disruptive and in unpredictable ways. Why not come out and say, “If the agentic software is good enough, we will fire people”? Answer: Being upfront is not something those who are not dinobabies do.
Stephen E Arnold, September 19, 2024
-
- Subscribe to Beyond Search
Feature archive
News archive
-
