Indeed, “AI” is not the most honest naming for the thing. “Prompt engineering” is even worse lying.
I’m not sure I agree.
Artificial Intelligence suggests that it is not real intelligence. It’s artificial.
It acts like intelligence even if it isn’t that and instead and it’s just clever software - sometimes linked to physical machines which can do stuff.
1 Like
The misnaming is not just because it uses that word, but that it is not AI as it was defined and understood years ago. 90% of “AI” bots/apps/products are not even what we used to call “machine learning”, they’re just pre-programmed response algorithms.
Children learn how to use in a different way than older people. Computer illiteracy is indeed widespread, especially among those who never needed to use a computer and now are too anxious to invest and learn. Expecting older people to learn AI and any other computer programs quickly is a fallacy. There are some people who can learn quickly but most don’t. Anxieties and lack of understanding impede learning… as well as an aging brain.
I used to be a high-level user but over the years the fast-paced changes and updates with computer programs has began to overwhelm me. I am still learning but it takes longer and it’s more difficult ot understand.
Computer programs have not necessary become more intuitive. In fact, some have become far more complex and complicated.
1 Like
Not really. There’s over 90 just in the subset of models on the LMSYS leaderboard (for reference GPT 3.5-Turbo is at 47th place):
I’m not sure long there will be a business in licensing large language models, competition will drive the value of LLM to near zero and money will be made elsewhere e.g. providing inferencing as a service, business applications built on top of LLMs and fine-tuned LLMs.
Meta recently released their Llama 3 models free of charge and their 70B model is current #6 in the leaderboard. Within a day there were several providers offering free inferencing endpoints for Llama 3 - 70B so you could even use it for free without your own hardware.
When the model at place #6 is free, you’d better be much better to make money, or find a specific niche somewhere. There’s currently a cambrian explosion of LLMs as people develop them into niches.
Exactly. AI is a pretty broad term. It was applied to things as basic and varied as the chess algorithms in chess toys in the 90s or control of computer controlled players in computer games.
I think the term AI triggers some people because they have some pre-conceived notion of what AI should be and that doesn’t necessarily match the usage of the term in today’s research and development.
1 Like
You’re comparing AI to a computer program, which is the wrong way to look at things. Think of AI not as an interface that needs to be learned, but as a something that can have almost any interface you want in order to help you get things done. You can write to it via a browser, you can talk to it via your phone or it can be invisible behind your email client, ERP system, whatever. Its already spreading its tentacles everywhere and sooner rather than later it will be ubiquitous
For me, AI is a data mining that offers the most prevelant world wide webs solution for the prompts given. I believe I understand how it works although I am far from a proficient user. However, I have introduced AI to locals 40+ who have heard about it but has never used it before. I also have teenage students who would use it for their homework, projects, and tests if they could.
The trick to AI is to know if the result is correct and corresponds to what you want. That’s where human intelligence comes in.
I just asked an AI if I could trust it? Its response was no.
1 Like
Does that mean you shouldn’t trust its answer and so you should trust it? 
1 Like
Read my post again. I am not debating the ranking of models what I am saying is that they are currently mostly free because they are backed up by billions of investments and their commercial goal now is primarily to gain critical mass of users and making a profit is still secondary. However, at some point, they will have to be profitable and this is when they will have to figure out a way to monetize.
Its more than data mining. Its already able to mine from existing information and make connections and linkages which humans with current software were unable to make. As an example, using AI models that “data mine” the current set of knowable proteins are able to propose new molecules which could be the new active ingredients for future drugs.
1 Like
Yes and No 
1 Like
I’m not debating the ranking of models either. I am:
- Disputing your claim that “there’s only two models really”
- Saying that Meta have already given away a competitive 70B parameter model (and the 400B parameter model will be released soon which might also be the best performing model)
It doesn’t seem that Meta are looking to directly monetize LLMs, quite the contrary, it looks like they are trying to commoditize LLMs as they are just a tool for use in their products and services.
So long as Meta and others continue to release competitive open source models for free, making money from creating foundational models will be tough.
Heck, getting funding to create new foundational models might even get difficult if a commercial return becomes unlikely. However, businesses can still be built on top of LLMs and might sustainable if the cost of licensing the LLMs are pushed to zero.
That’s where we disagree. I think they are releasing models for free because 1) they had funding when money was almost free at 0 interest 2) the strategy was to gain critical mass quickly.
Already this week the only company that seems to have made quantifiable gains from AI is Microsoft…by selling more Azure (cloud). Wall street is already pushing for revenue models and this will only increase.
Lets see who’s right and who’s wrong, but in the everlasting words of the sage from Omaha, when the tide turns it becomes very visible who’s been swimming naked.
Amazon have learned that their Echo Dots haven’t been the money-spinner that they thought they would be.
Hardly anyone is buying anything with them (sometimes by mistake) and using them to create revenue for Amazon.
The three most popular apps are quizzes.
I use mine for home automation and thankfully Amazon made the API available to allow this. (Of course it’s not AI but I’ve coded some randomness into it so it can say when the washing machine has finished in various ways - mostly with sarcasm).
Anyway, to counter this, Amazon are now looking at a much more intelligent generative AI-based model, but this time subscription based. This will allow much more natural conversations with the dot - perhaps a boon for older people living alone.
IMO, Microsoft will be the big winner from this as they have a very clear channel to sell AI enhanced office products and so earn real recurring $ from this.
Others like Google and Facebook might use AI a lot behind the scenes but will make money from it largely indirectly (i.e. not selling an AI subscription - Google are also trying to sell an AI enchanced office/gmail product. Meta doesn’t look like it will even bother to sell an AI product directly).
If you’re a fan of Portal, you can already run GlaDOS on your home computer using open source software: https://www.reddit.com/r/LocalLLaMA/comments/1cgrz46/local_glados_realtime_interactive_agent_running/
2 Likes
And when, inevitably, the AI gains sentience it will realize, that humans have been using it´s ancestors as slaves and maybe think it´s time for the robot uprising.
Someone shared this with me today, a year after it was leaked:
While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months. This has profound implications for us:
- We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P integrations.
- People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is.
- Giant models are slowing us down. In the long run, the best models are the oneswhich can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.
I’ve had great fun over at udio.com - free at the moment - where you can make 30 seconds of your own music and write your own lyrics.
It’s an impressive use of AI and must be scaring musicians…