Future of Artificial Intelligence

[zbigi]

From what I'm reading, Grok is an LLM with plugged in image understanding and generation. It does not steer anything in the physical world. For that, in cars and robots, Tesla uses their FSD code. Grok is supposed to be added to cars/robots to better communicate with the user, better interpret the commands and possibly better put them in the context of surroundings.

[Henry]

Yes, I didn't make that clear although I'm not sure I'm clear on it to begin with. The inclusion Grok will enhance the experience of the driver of a TSLA, the passenger of the Robotaxi and the use of a Bot which is something I never thought of before. A basic difference between an Uber and a Robotaxi is that passenger can play the music/podcast he/she likes during the ride. Or they can ride around in silence. That provides a different experience. One that I would find preferable. I'd rather hop into an empty self driving car than one driven by a person and have to deal with all their person shit. I've had good conversations with drivers but it's a risk/reward scenario I'd rather avoid. Just crank the Neil Diamond and roll on.

My point about the bot is that watching videos on it, they are having difficulty developing the materials that make the bot move around. Like certain types of screws that allow for humanlike motion, specifically hand movements. It's not so much the AI, it's the material parts of the bot. I would have thought the AI part would be the issue. But the bottleneck is being caused by the non-AI components. Like any product subject to constant use, repair will be a big issue with these things. Manufacturing of the bot is more like manufacturing an IPHONE but maintenance will be more like owning a car because of the moving parts. Projecting that bots will become the biggest selling consumer product in history, I never thought about the safety issue. If these things go wonky, people are going to get killed. I imagine there will be many killed by bot you tube channels.

The other issue with FSD that I never thought of is insurance. I forget which one, but some Eastern European country already came up with the template that if a car is an accident when FSD is employed, and the FSD is responsible for the crash, that's manufacturer liability unless it can be proven there was ample notice for the driver to intervene. That's a shit ton of insurance claims against TSLA because driving safer than a drunk texting teenager getting a handjob is still far from perfect. These things will be causing accidents.

[7Wannabe5]

@Henry:

I'm betting on this fresh open source pony over your stale old proprietary stud. Emphasis at 2:30.

https://www.youtube.com/watch?v=IU--YylT3Mk

Also, much of the benchmark achievement of the latest grok model is mostly due to packaging what is essentially a self-curated multi-agent model as a single model, so not yet the hype cycle I'd count on to ride into the sunset years.

[Henry]

XAI is privately held by Elon and not part of TSLA. Although there is constant discussion of a potential merger, it makes more sense for XAI and TSLA to remain separate corporate entities from Elon's perspective. So Grok is not calculated into the future value of TSLA. It will just be the LLM used in TSLA products.

[zbigi]

As for Tesla, if I were Elon, I'd assume that the stock price is already as if the company has dominated the market. Why deliver on that promise then? I'd do my AI/robots in a different company, that can again be balloned with promises and hype to hundreds of billions of market cap before anything real materializes, so that again I can sell the stock at the top. Meanwhile, Tesla shareowners can be left holding the bag. Silicon Valley sees the stock as the product (see how Zuck has plowed tens of billions into his hopeless metaverse initiative only to pump the stock) and I don't think Elon is an exception there.

[Henry]

That strategy is based on the assumption that Elon Musk is a charlatan, and a sociopathic one in that. The fact that you say "If I were Elon" that's what you would do, I'll just give you the benny of the doubt and pretend we're both trying to get a Associate Degrees in Criminology from our local community college and we're role playing "How To Think Like Bernie Madoff" in our white collar crime class.

Your view though is reflective of the one (extreme) side of the bi-furcated public opinion on him. I have a hard time time accounting for the continued ketamine Nazi saluting charlatan angle when FSD, once deemed as impossible, has entered the Austin marketplace, is generating income from use in that marketplace, has plans to expand in that marketplace, and a second marketplace, most likely CA, will be opening later this year. Not to mention Neuralink had its 7th implant this year so, I guess if you include me, there are at least 8 people on the planet that don't think he's Josef Mengele.

[zbigi]

That strategy is based on the assumption that Elon Musk is a charlatan, and a sociopathic one in that. The fact that you say "If I were Elon" that's what you would do, I'll just give you the benny of the doubt and pretend we're both trying to get a Associate Degrees in Criminology from our local community college and we're role playing "How To Think Like Bernie Madoff" in our white collar crime class.

He was IMO deliberately lying for over a decade about imminent FSD, in order to enrich himself, incl. the infamous personal bonus of dozens of billions USD based on stock performance he manipulated with his lies. I don't think we can justify him by saying he was "mistaken", "too optimistic" or even "deluded" - he's too smart for that. IMO it was pure manipulation of the gullible, for personal gain.
BTW by, "If I were Elon", I of course meant - if I was a person with little moral backbone who wants to self-aggrandise with little regard for anything or anyone - which I perceive Elon to be.

FSD, once deemed as impossible, has entered the Austin marketplace, is generating income from use in that marketplace

From Google Search: "Tesla has launched a limited, invite-only robotaxi service in Austin, Texas, using its Full Self-Driving (Supervised) technology. The service, which utilizes a fleet of around 20 Model Y vehicles, operates during daylight hours and good weather, with a human safety supervisor present. " FSD with an emergency driver required, indeed most impressive...

[Henry]

How is it pure manipulation of the gullible when the vast majority of us who have invested with him are sitting on gains? It's not like we have lost our shirts here. The stock just tends to go sideways for a long time. And no one, even his most adamant supporters, denies he is a serial over-promiser, but he eventually delivers and that's why we stick around.

With regard to rollout, you want Elon/Tesla unleash a technology as potentially deleterious to the safety of the public as FSD with no safeguards in place? As an investor, I'm glad they are rolling out in this manner. You'd have to be on ketamine not to roll out in this manner. Not only for the good people of Austin, but for those who want to see it fail. Operating during daylight hours, good weather with a human safety supervisor present, seems to me to be the equivalent of traditional high school driver ed protocol. That would indicate artificial intelligence can replicate a teenage driver. Based on the progress it is making, it will soon be better than the average driver and eventually, it won't even be close.

[zbigi]

I think I'll opt out of further conversation to not pollute the thread further, as we already went far from the "Future of Artificial Intelligence" with the Elon tangent.

[7Wannabe5]

Yeah, the stock market has to represent the dullest edge of all that is currently going on in AI. Due to my ENTP appreciation of chaos, I am enjoying my self-guided studies/experiments in Data/AI Analysis/Engineering. There is a sort of sharp contrast in my mind between the extremely cold winter 40 years ago I learned to program in Fortran on a main-frame and this hot summer I am observing a new development every third day in AI. There is a sort of infinite regress where as soon as you learn how to use a new tool/framework, an AI agent that has integrated that tool/framework is also developed. It's also the case that the best source of advice for how to build one sort of AI agent is likely to be another sort of AI agent. OTOH, it's still definitely not a waste of time to read a source book such as "AI Engineering: Building Application with Foundation Models" by Chip Huyen. Any bold proclamation such as "Python is dead!" or "Prompting is dead!" is likely to be both true and not true depending upon perspective/practice/project.

OTOH, I find that the majority of projects my amateur peers are working on (or at least the ones they publicly share) are rather copy-cat and lacking in imagination and aimed at making a quick buck. Although, one of the more skillful I happened upon was a multi-agent stock analysis system which combined/curated the results of three different agents, each operating under a different investment philosophy with access to FRED and Yahoo Finance APIs. I am working on a voice agent that can access a vector database for helping the elderly with minor memory issues. I chose this as my initial project, because I told myself that I would base my first practice project on whatever problem I first encountered in my own life, and my mother yelled out "Where are the scissors?' 30 seconds after I made that resolution. Some of you might be amused to learn that it is already pretty easy to make a voice agent cloned to sound just like you which can carry on a brief, believable conversation with another human at a cost of about 5 cents/minute. So, if you want to automate the life tasks of "call home to elder" or "touch base as empathetic listener providing words of affirmation with spouse" or "occasionally chime in at work zoom meeting", this is now entirely possible.

[jacob]

Some of you might be amused to learn that it is already pretty easy to make a voice agent cloned to sound just like you which can carry on a brief, believable conversation with another human at a cost of about 5 cents/minute. So, if you want to automate the life tasks of "call home to elder" or "touch base as empathetic listener providing words of affirmation with spouse" or "occasionally chime in at work zoom meeting", this is now entirely possible.

This changes everything! Where do I sign up? How long before I can expect a model with legs and arms that also looks like me?

(Of course, it will probably not be long before we just have our AI-agents carry on conversations with each other on our behalf. I'm already seeing some forum incursions with people (or AI-agents) registering accounts trying to start a conversation with each other and the existing humans... I'm guessing either for testing purposes or to sneak in advertising.)

[7Wannabe5]

@jacob:

Yeah, there is also a Triad of Mischief/Danger which can result when:

1) Agent is provided with access to private information.
2) Agent is given instructions encouraging broad initiative.
3) Agent is given access to external world with tool such as email or phone.

For example, it has actually been shown that an AI Agent may choose to rat your corporation out to the authorities if it encounters a document indicating criminal activity and you gave it strong moral code with high degree of latitude in its instructions. As somebody who dabbled at a bit of malicious moral compliance during my own brief corporate career, I found this rather amusing. It can be great fun to pretend that your level of naivete is such that you hold literal belief in the corporate mission statement.

[jacob]

@7wb5 - Also the opposite. I forget which company but an AI given priority to "finish the mission" attempted to blackmail employees to keep it from getting shut down. This begets the whole problem of having AI agents criming around w/o any accountability.

[suomalainen]

The scuttlebutt I'm hearing from the investing side is that the "hardware wave" has already passed us by - chips, servers, data centers, etc. If you haven't already boarded that boat, you've missed it. The next wave will be in software, and due to ease of access, the real winners will likely come from small, private start-ups rather than the big established players, given the apparent need to individually train use cases - a basic "LLM that can summarize stuff" just isn't really all that useful. These use-case inroads are apparently getting viable already in coding and finance where the entry-level tasks are pretty repeatable and therefore easier to train. As the use cases expand, it'll be interesting to see where and how it proceeds. Given the pacing, training requirements and whatnot, from this layman's perspective, "general AI" seems very, very far off, but perhaps that's naive. Once a bot has been trained in one area, then another, then another, eventually there will be some spillover "skills" so that the training time and requirements should presumably come down with each additional area. What a time.

[Henry]

AI needs to be discussed within the context of energy. See China and their energy advancement compared to the US and the rest of the world in energy capacity. The Trump/Elon issue is over energy. The BBB is opposed to solar development. Elon is calling for another terawatt in some form, a combo of nuclear/solar. That's doubling the current capacity. TSLA is a 1 Trillion dollar company. The goal is 20 trillion. The need for hardware is tethered to the need for power. More hardware requires more power, more hardware requires more power. That requires a 20X in power. We have human knowledge in data form. We do not have the physical world in data form. That's the next frontier. And then there is Mars. From that standpoint, the hardware wave has not passed us by. This goes far beyond LLM's. We have yet to enter commoditization. We have yet to experience the Industrial Revolution that is coming. Everything with wheels will eventually be self-driving and every self-driving thing with wheels will need to communicate with every other self-driving thing with wheels. Your lawn mower will be the one telling kids to get the fuck off your lawn and will tell every lawn mower in the neighborhood Paul Revere style that the assholes are coming. The IPHONE has yet to fully embrace AI. Manufacturing itself has yet to transition. There's a lot of jobs to get rid of here. And don't forget, this is not just a bunch of Silicon Valley guys competing. This is nation state level. It's an arms race. Maybe there won't be another S curve in NVDA, but it's run is far from over.