I have worked in many organizations over my 30+ year career and one thing I learned is that the modern corporate ecosystem, on average, encourages the incompetent (more or less) to rise in the ranks of management. I know a lot of you have also learned this. The way to succeed is mostly just make sure you can blame others for failures and don't get in trouble. Above all, their like things that save them time. AI looks like it can do that for them.

However, I've already seen in my own work place and the work places of others I talk with situations where developers were fired with AI to replace them. It failed and they then started trying to rehire the lost positions. Sometimes, they really are just looking for reasons to get rid of certain individuals. The executives are those often more interested in firing to save money. In those cases, it's usually more random or whole departments.

This all said, I develop AI tools and refine models all day long. I also use AI to code for me. I do find it useful but it took me a lot of time and work to learn how to use it effectively. It really does require skill and knowing where and how LLM models screw up. For example, don't just tell them what you want have them make it. Give detailed descriptions of the tools, protocols, coding style, etc. Then design the tables/columns, the API calls, etc. Design your software at each layer and have it build each layer and test and debug each layer comprehensively before moving on. Then, in the next layer, tell it not to edit the other layers. Or at least, tell it not to modify the regression tests for the previous layers. It will give up on bugs and try to hide them or belittle their importance, etc. Force it to fix them... keep trying until it gets it.

I wonder if a new AI-enabled Wikipedia app might help? I will need to look at what the Wikipedia public API provides but an idea I think would be useful is a Wikipedia research and production tool. That is, to build an application that helps research a given topic and compile a book that includes the relevant Wikipedia page content along with a synthesis or multiple synthesis bringing them together. Obviously, we'd use an AI model to build those synthesis. The output could be online readable content, PDF books, or other kinds of reports. I don't know--just starting to think about it.

One thing I have long wished we had was a news site that aggregates facts, corroborates them, and synthesizes them over time. So news will not be a single moment in time but facts given along side various permutations of interpretations thereof based on relevant history. In fact, this is my advised new model for news media. They could offer paid subscriptions to such a service. It would be very valuable for many kinds of research.

While I am no advocate of the absurd complexity of npms (though I maintain a few) or node's method of dealing with async, I personally find the JavaScript has a superior method of Object-Orientation and is a more practical and much faster executing language over all. I know--there will be strong opinions on this but JS is much faster and simpler. I think there is a fundamental difference of opinion as to whether or not your language should control you (Java/C++) or you control it (JS/C). I aspire to the latter (and understand many others won't).

Imagine the power, performance, and low memory foot print of your PWA applications with Javascript replacing the Java layer over Linux in Android. Imagine how much faster and easier to build and maintain (if you avoid bloated frameworks like React and Vue).

This is the phone I would want and the phone that could last more then 3 to 5 years.

Look at the long list of utterly absurd Salesforce patents. It is hard to find a single one that could be even arguably unique. It is mostly just a list of common things any and typically every CRM does. I looked through these long ago and saw--at a glance--what a joke it is. Yes. The patent system is near completely dysfunctional, even after so many reforms. Nobody in government seems to care.

Just understand this: A major part of any business plan is legal. Contracts, leases, or patents are not, in and of themselves, any form of protection. It's about what you can defend or strike back with financially or the realistic threat thereof. Legal documents like these are often just a way to sucker people into deals thinking they are two-sided.

These thing rarely even continue to a judge's ruling. It's about who can intimidate and cost the other more in the process. It's about bleeding your opponent.

1st Bubble -- Yes, there is a lot of money invested into so many companies rapidly building AI products and, mostly, retrofitting old things. Obviously, they cannot all be winners.
2nd Non-Bubble -- In spite of a lot of failed projects, there will be more project and there is already too little infrustructure for the demand. For example, Microsoft 365 Copilot is clearly a weak GPT model but vastly under powered (easy to notice if you use it). The first wave of AI companies will have some winners and a lot of loosers but there will be wave after wave after wave because there really is so much unexploited potential.
3rd Bubble -- Indeed. Most of these products are boasting capabilities they don't really have. Maybe they do, marginally. For example vibe coding. It looks so incredible but there are numerous problems making it largely useless in real world situations. For example, it overstates what it does and often cheats (by hard coding or pretending). Second, it breaks down as projects become larger or more complex. The context windows just are not big enough. These tools can be used effectively but it takes a lot of time and practice trying things to figure out how to do that. I am aware of multiple companies that fired developers thinking AI could do their jobs only to have to hire them back, later.

Initially, I found the same in myself--a real degradation overall in my productivity. I am a software Engineer. It has not been easy learning how to use generative AI to actually increase and improve productivity. At first, you think it can do almost anything for you but gradually over time realize it greatly over-promises.

Overall, the key is that you need to remain in charge of your work. It is an assistant that can be entrusted more or less to small tasks with oversight, at best. For example, frame out your project and clearly define your rules and preferences in fine detail. Then..

It's good at:
- Researching, summarizing, and giving examples of protocols, best practices, etc.
- Help you identify considerations you might have overlooked.
- Writing bits of code where the inputs/outputs and constraints are clearly defined.

It's bad at:
- Full projects
- Writing anything novel (it knows common patterns and can't work much beyond them.
- Being honest and competent -- it cheat on writing code and writing tests for that code; when you catch it red handed, it will weasel its way out.

The bottom line: you are in charge. You need to review what it gives you. You need to work back and forth with it.

Also -- I am still learning.

--Matthew

Contemporary AI, and particularly generative AI, has extraordinary potential but can only fully do a job in a small number of roles. Those roles are where it is used as a tool.

What I mean by a tool is that it is driven by assigned goals rather than free will value judgments. This makes it useful as an assistant to a human, at best. An additional limitation is its inability at fundamental creativity/innovation.

Intelligence is that ability to solve problems. A human mind is centered around something more universal that that. A human mind makes judgments in exercising free will. We freely derive options, weigh those options against each other based on the sum of efficacies (likes/dislike amount) multiplied by likelihood (likely/unlikely amount) of the foreseeable results. This allows us to think outside the box by throwing away the goal as stated in preference for solutions or goals that better meet our interests. It enables adaptation.

Further, contemporary AI is not capable of analogy except in very meager ways. Analog is the mental imitation of one thing with another where each share characteristics. By noting the solution to a problem requires one component that can be fulfilled by any of a set of options but none are available, we can hunt for something novel that shares the same characteristics. This is substitution problem solving. It is making new connections. This is fundamental creativity/innovation.

There is literally no reason for youtube, facebook, X, tictoc, instagram, netflix, etc.
All of these could be better replaced with open protocols for which you use your own choice of client applications or make new ones.

For streaming movies, for example, imagine where movie studios just make their movies available and set their prices and conditions. Similarly, consumers specify the prices and conditions they are interested in and the matches are made automatic. For example, I want to pay $30 per month for access to 20 scifi, action, and action-adventure movies and series, randomly selected at the beginning of each month, preferring newer, for the month. Have an option to specify which to keep and when to remove.

The protocol needs rich criteria and logic building but this can be done. Studio's made need to buy server / storage hardware or pay third party for that but not a big deal. Consumers could create the plans they want and the market would show them what they'd get for what they want and want to pay for it.

Also -- why have a single company in charge of tweets? There is utterly no reason for any kind of proprietary control over social media. There should be good controls over how one chooses to accept/block content of interest to them. We could implement criteria for judging the legitimacy of posts to fight fake information. For example, are their primary sources? Crowd source the validation of this.

Indeed, Intel simply made bad bets on the future and relied on milking while coasting on its past. This is an obvious move for short-term profits over long-term potential. On the other hand, I do see a way forward that could work for them -- be early adopters of RISC-V. Yes, it will be a commodity architecture but Intel could take and hold the lead by being first to mass production at high integration (even if not the very highest) and through development of custom extensions for edge parallel compute. That is, hook future software on proprietary extensions for running local AI. NVIDIA will lead with high end AI such as for building and fine-tuning models but Intel could jump to the lead of running those models on local devices in about a year's time. In two to three years, Intel could be leading in desktop and portable devices with this technology. This lead could be held.

Boeing, however, has manifest problems all across the board. This is due to adoption of an extreme stupid management system and, further, double-down on it after the 747 Max disaster. This destroyed McDonald Douglas who then merged with Boeing leading to Boeing adopting its management practices. There are no nice words appropriate--this was utterly stupid then and stupidity continued and continues even now. It's tragic. I loved Boeing, growing up, and have been forced to watch it self-destruct in slow motion over the years. The only way to save Boeing is to let it fail and allow it to be purchased and rebuild with those who put the engineers back at the center of the company.

In both companies, it is critically important to put engineers back at the center of the company (development and manufacturing) and to isolate and insulate them from the business around that. The engineers and even the assembly line works need to have ownership of their processes. They need the right to say what is practical and not and to provide the schedules in consultation with marketing and sales and the excutives and not the other way around.

China has already amassed a vast mix of different kinds of drones with autonomous modes--and in very large numbers. Ukraine is also making increased use of autonomy in their drones, as a way to get past radio jamming. In the case of this immediate need, they don't have much time to train their drones of differentiating between friend and foe. It would be a GREAT benefit if Silicon Valley would stop the idiotic debates and start developing the kinds of image processing and AI that can make the best possible decisions in combat scenarios. Otherwise, only the bad guys will (and are). No matter how much we want to live in a peaceful world, the realities cannot be ignored. Russia and China, in particular, have been using trade and economic development for the purpose of preparing for military conquest. The old ideas that a growing middle class would lead to a populace demanding increased civil liberties and a share peaceful world was a good idea, 20 year ago. However, it has clearly failed in authoritarian regimes--and is backfiring.

If China were to begin an invasion of Taiwan with an infiltration and release of drone swarms, they could quickly take down Taiwan's air and coastal defenses allowing the Chinese air force and Navy access. Taiwan has no realistic defense against drone swarms. Neither do we. Our largest adversaries already have them. China has been building up oil and food reserves. It's been my hope that China would see invasion of Taiwan as foolish after seeing what happened in Ukraine to Russia but it seems, rather, they've earned a ton of money selling drone parts to both sides and have learned a great deal from Ukraine's drone innovations, amassing a vast quantity and diversity of their own--with autonomous mode.

AI that can search, cite, and link while integrating information by relevance to the query is obviously vastly superior to Google search.

Similarly, however, AI that gets to know the user and his/her needs/interests and projects could be a bastion for business IF the AI company charges to list their products. The key to win would be to offer more information about the products thereby enabling the AI to determine what products meet the user's actual needs.

When I say products here, it also applies to services.

I had this idea long ago with my old chatbot engine and software. However, I never did it. This was prior to chatbots becoming popular and prior to generative AI becoming popular... Today, this could be even vastly better than what was possible then. It would not be difficult to implement.

AI generative models are not an end in and of themselves and are highly educated though not very smart. However, this as a component of larger products has extraordinary potential. So far, companies have been rushing to implement it into their existing products and that is typically less impressive than the hype. However, architecting new products with AI at the core is a different proposition. It may take a while for big new things to come out of the woodwork but the material now exists to make it so. Imagine not AI features added into software but AI driving the software. Basically, it will make things simpler and easier to simply talk with your software to get it to do what you want or show you how or pull what data you want from it. No more need to learn an application but rather focus on what you want to do with it. I think ChatGPT, Claude, and others may not fair so well in the years to come. However, the commodity models from their work will endure and arise in new forms. Why? Because there is so much we can do with them as integral tools to new things we design and build.

We notice that the legal argument against AI by writers and artists is that it is a form of data compression and past cases established that data compression and copying the compressed versions of data, even if lossy, still constitutes copying.

I disagree with this argument used against LLMs. However, it is a pretty good argument. Indeed, it amalgamates information and can recreate similar works to what it has read. It does so in a very similar way to a human mind, though, and yet we allow humans to do this. You are allowed to paint something similar to the Mona Lisa, after seeing the painting. So what is the difference?

I've heard only one argument as to what the difference is--that it was viewed by a person. Legal personhood, such as that granted to a corporation, requires that the agent be (1) reasonably able to accept legal responsibility and (2) can be held to account when it fails in that responsibility. This is why a dog cannot be called a person and the mentally disabled are sometimes required to have a guardian. The argument was used against granting legal non-human personhood to even highly intelligent animals, such as dolphins and animals that are very genetically closer to us, such as chimpanzees and arangutangs (even closer genetically than some born from human parents with genetic mutations).

So I believe that an AI capable of being given legal responsibilities and being held to account for them should certainly be able to view writing and art and recompose its own based on what it learns from those experiences. In this case, there is likely no other way to argue it is copying unless it actually does copy in a similar way that a human might copy. There already exists a theory for giving legal personhood to an AI agent. It involves incorporating the agent, setting up another corporation that owns the agent corporation and then transferring ownership of the second corporation to the first. This gives the agent legal ownership of itself in addition to recognition as a person by means of already established law.

The results were obvious. This was an example of bypassing education with AI, not educating with AI. I spent a number of years developing techniques to teach using Chatbots. These could be powerful educational tools even before transformer models.

Here is some quick advice:

1. Use the give-take approach of giving a fact or problem-solution and asking a question/giving a problem, in sets.
2. Employee timings based on Graduated Interval Recall where if the answer isn't found within that time frame then give it to them. This is useful where the problem can be broken up into small enough parts.

With LLMs, you can add very rich question/answer/explanation scenarios, also. I had this capability but it is much richer than before. I like to ensure memorization through re-validations on a timer. First, it's 25 seconds, then for the next time, just multiply by 5, and continue the longer the better. This creates spaces between to add ore content. There is a good neurological basis for this. Axons on a neuron will split after 5 pulses not to close and not to distant from each other in time... It would take a while to explain in full.

A journalist soliciting classified information is already committing a crime (if it wasn't then it should be). What's worse is that Assange willfully released it knowing full well that it would lead to the deaths of various listed informants--and it did. That is murder, negligent at best, but he did so with clear anti-American intend and hatred. This deserves no sympathy.