I've been super-busy with travel, work (starting a new job!), and other real life stuff. I've been meaning to put together a bit of an overview and some thoughts on Generative AI....I'll have this post now and another later regarding my personal thoughts on AI as well as the various projects that I've been playing with "Real Soon Now". I'm talking here specifically about Generative AI, which is a subset of AI, a general, almost just buzzy, term that generally refers to various classes of software where computer models have been trained instead of programmatically determined. This can mean anything from Machine Learning (which has been in products for 15-20 years) to Artificial General Intelligence (AGI), where computers actually think, which does not exist yet.
The below (although a whole lotta words) isn't exhaustive on the state of play or the criticisms by any means, and the technology and the business and social movements surrounding it change daily.
The state of play in GenAI
The GenAI technical landscape is getting more intense, both in terms of GenAI complexity and sophistication (growth of parameter sizes and larger context windows) and the move toward agentic AI. Generative AI technically has been around for decades. Many of the models being used have been around for a long time, but commercial projects weren't really possible without the gigantic amount of training data that exists today - volumes of data on the internet including digitized books, social media interactions, photos and videos on various platforms, digitized music, etc. It's only now that companies can take this data (sometimes through what many would call unpunished theft of copyright) and train their current and future models on it to build their models for commercial use.
GenAI is getting to the point with both text-based and photo/video processing that it's increasingly difficult to tell the difference between being 'live' and 'memorex'. Agentic AI is the new hotness - there's new tools that can be run on a computer or a network of computers that can be tasked to do various things - manipulate files, run various online services for you. Really anything that you give the tool access to. Some tools you can give a credit card number to (scary). These tools like OpenClaw or Claude Code won't only help you build out projects, you can give it a prompt to do something and it will find a way to do it for you (this is agentic AI).
On the business front, it's a land rush. The draw of the 'new hotness' of GenAI has made a lot of companies AI-first - many business leaders feel that if they can grow their business (or cut staff) by utilizing tools that can produce more and better code, faster analysis, cleaner documentation, etc, now is the time to do it. They don't want to be left behind. Similarly on the 'supply' side, there's been a lot of economic change surrounding GenAI - almost half of the S&P 500 index value in the US is made up of GenAI stocks, which only represent 8% of the companies in the index. Three of the biggest movers in GenAI have either IPO'd or will IPO soon, all of them approaching or exceeding $1 trillion valuations. There's also literal land rushes to build the data centers necessary to train AI models and serve traffic, especially in the US.
Criticisms
There's plenty of dangers both from GenAI itself and from the impacts coming from the growth of AI:
Social factors
While GenAI is getting a lot better at being an inference engine, that's still what it is. It's not a thinking engine, and people have found it dangerous to think otherwise. In the case of GenAI chat, it doesn't give you the truth or the right answer. It infers what the answer 'should be' based on what you said and what's still in its context window. Sometimes the answer it provides is brilliant or at least thought provoking. Sometimes the inference is flat-out wrong for any number of reasons, which is often called a 'hallucination' but is still frankly just bullshit.
The problem here is many use GenAI as an ultimate arbiter of fact, or believe they are 'connecting' with GenAI as if another person. The same kind of people that do web searches on the internet and blindly agree with the results given to it are the ones using GenAI without critically thinking about the information given, asking clarifying questions, or fact checking its responses, an enhanced 'filter bubble' way beyond customized web page results from search engines and social media networks. There's people that have been led to financial ruin or further down a mental health spiral including, horrifyingly, "suicide by AI" because of lack of controls in GenAI and those same people unfortunately trusting what the software would tell them.
GenAI, like any other software (or indeed, any product that's has a sophisticated design) is going to have the same goals and biases as those developing it and especially the companies that are utilizing it to build a profit model -- In Google Gemini's case, for example, its chat functionality ALWAYS ends each question/response 'turn' by asking a question to keep people engaged (and gather more data to train its models).
Economic factors
It might crater the economy There's a VERY good chance that large-scale GenAI (OpenAI, Anthropic, etc) might not be a success economically. We'll talk about data centers more in a little bit, but the newer larger GenAI models are insanely hungry for large amounts of very expensive processing, and these are usually provided by custom processors made by Nvidia (which has become one of the world's most valuable companies in the last few years). But GenAI companies at this point are providing access to their platforms at a price that doesn't get anywhere close to the actual cost of providing the service, much less profitability. They are relying on the standard model of Silicon Valley fund raising, but if it fails the amount of money involved here may well cripple the US economy.
Let's take a quick dive into how Bay Area startups work - companies start up with small amounts of seed money/angel investing because they have a good idea. Even if they don't have a product yet, if these companies can start producing operationally, they are given more and more 'rounds' of money. These early investors invest because they want to make a (large) multiple of their money -- an "exit". The goal is to either:
-
Door number 1: The company is either extremely profitable or is on the roadmap to profitability and they go public (IPO), giving investors their 'exit'.
-
Door number 2: The company gets acquired by a larger company at a multiple of what was invested, also giving investors their exit.
There's obviously failure modes here too - companies fall apart because of mismanagement, they can't get the right 'hook' for their product, they are entirely fraudulent (see Theranos as a prime example here). The worst one is when a company is way over valued for what their product actually is. If a company has a decent product and sells investors on a dream that the product is a lot more than it really is, they can get pre-IPO investors to invest WAY MORE money into it than the company is actually worth, and when those companies try to go public investors get a giant haircut on their investments. WeWork is the prime example of this situation, and was so recent I'm a little surprised that investors still have stars in their eyes regarding GenAI.
We're in what's probably the third major paradigm of internet startups now, the first being your Apple Computers or even Dell and the like - they were profitable companies and had a "killer" business model in place before going public. the second paradigm was internet startup land. Most internet startups up until now, some successful, many many are not, build on the idea of a minimum viable product and then iterate from there, see what their market differentiator is, and focus on that. Some companies do extremely well with htis model. Companies like Google were somewhat popular and had some revenue trickling in, but even with them their monetization engine (AdSense) was already in place when they went public. Pets.com on the other end of the spectrum is a posterboy (posterpup?) of startup over-reach and was one of the more disasterous collapses during the 2000-01 dot-bust, they became a household name due to great ad campaigns and went public (giving their early investors their exit) but their costs outstripped their income and eventually folded.
So the paradigm in the second case is essentially to fake it till you make it. Sell a product at a loss until you 'win': come up with a sustainable killer product, achieve market dominance, succeed operationally so your costs stay down. One of those things. And companies that have a 'burn rate' of a few million bucks a year in AWS costs can do this because if they have a good business model they can continue to attract investors until they 'win'. But this model of fake it till you make it has two really big downsides, especially when it comes to GenAI:
-
Really Big Downside 1: Investors REALLY want their exit - there's a lot of other really big problems with this business model that I'll probably rant about later but for Gen AI specifically, this means investors lie to the public, will twist the arms of government officials to get approval or even tax rebates for data centers, etc. A good example, there's a lot of controversy around city and county leaders in the US that are giving sweetheart deals to get data centers built that the residents of that area want absolutely nothing to do with. They'll engage in tricks where they fold really toxic companies into successful ones -- as in the case of xAI and the social network formerly known as Twitter getting folded into SpaceX in order for Elon to pay back the Saudis and Larry Ellison for bailing him out of his disasterous Twitter purchase (sorry for that aside). If you look at the circular investments propping up companies like OpenAI, where Nvidia is investing heavily in them on the promise that they'll build a bunch of data centers and buy Nvidia chips to run in them, you begin to wonder how this is still even a thing. If they don't have a financially sustainable business model, they'll (metaphorically or not) lie, cheat, and steal in order to make one.
-
Really Big Downside 2: The scale of the businesses we're talking about is truly breathtaking. This isn't Pets.com, who raised $~80 million on their IPO. Both OpenAI and Anthropic are looking at valuations at $800 BILLION dollars each. If/when these companies go public at that valuation they'll be in the top 25 of all public companies in valuation instantly, without being anywhere near profitable.
All of this to say that the previous paradigms of the tech world don't yet apply here, and with the current political culture of lawlessness opens the door to some really weird and potentially awful consequences.
It's already disrupting tech hardware Gen AI is currently causing major disruptions in technology supply chains and prices. The amount in hardware sales has been great some vendors -- especially Apple, which is making bank on businesses and hobbyists that are buying powerful Mac Studios or Mac minis to run local LLMs. But it's also been brutal for cost of memory and other computer components, which have risen dramatically. Apple was able to stave this off because they had existing supply chain deals but those ran out and they just increased prices by $500/system in many cases.
It might cause mass unemployment If Gen AI IS a success, it's going to cause widespread job loss, and that's something even those who are AI cheerleaders have said out loud, and have also admitted they have no solution for it. Some talk about Universal Basic Income (UBI), which generally I think is a good idea but those who have brought it up (Musk, etc) are coming at it from a billionaire's perspective and not really looking at allowing for self-sufficiency.
It might break career advancement Beyond tossing millions of people out of their jobs, GenAI (if the economics work) is going to cause really weird distortions in the human workforce. Even if issues regarding data accuracy can be fixed to the point of acceptability, it's going to at least initially get rid of a lot of entry level workers. And since senior level workers became senior because they were once entry level it's going to cause a lot of issues.
People in my job (Senior IT Nerd) are only good at their jobs because they've had years of experience, seeing how to deploy things, how to troubleshoot breakages, etc. GenAI is a bit of a game changer for people at my level because we can build various tools (skills, MCPs, etc) that do a lot of the dumb stuff for us, making the job both more interesting and faster to get results, but you have to have the skillset coming in to be able to tell a GenAI tool what to build, and you have to know when GenAI is building the wrong thing. Without that, you get into a paradoxical situation where GenAI has made junior IT (and regular engineers) redundant, but once senior engineers retire there's nobody that will have the skillset to do the job anymore.
Environmental factors
Where to begin? Gen AI requires a giant amount of energy, water, and other resources for the construction and usage of data centers. Data centers also gives off a lot of heat. Because corporations are... less than ethical sometimes? they'll work with the same local officials that got them the sweetheart tax rebates to avoid environmental assessments, data centers are going into places where energy and water are cheap and the deals they are signing with local utilities are making water and energy more scarce and higher priced for residents.
When you have super-high energy needs and only a certain amount of power generating capacity in place (and a US government that hates renewable energy), GenAI companies are turning to fuel that is really awful for the environment. xAI for example built an unpermitted power plant to power its data center near Memphis that burns methane, and other GenAI companies are renting time from xAI (because they can't find enough data center space to handle their growth).
There's also the recent news story that Meta as part of their construction of a data center dumped bacteria into a local water supply, causing the water treatment plant to have to scramble to handle it before it made the local population sick.
Essentially the race is to get to Artificial General Intelligence or alternately get to a sophisticated enough version of GenAI that can solve all the social, economic, and environmental problems that it causes. What's your bet on this?
Stay tuned for part 2 where I talk about my thoughts on and experiences with Gen AI.
Here's some resources if you are interested in learning more:
If there's one article about Gen AI that I recommend people read, it's this one, where Om Malik says a lot of what I said, just better.
Agentic AI Foundation CTO Manik Suritani - who owns the pipes?
Financial problems - Ed Zitron's site and his Bluesky account is a good follow as well.
Chris Hayes' podcast Why Is This Happening podcast did a series on GenAI recently.
Recent Dispatches
- Apr 16 — Platform SSO for macOS - An Overview
- Apr 12 — The Tax of Inertia in IT
- Apr 08 — First!!