AI - Artificial intelligence (2 Viewers)
- Thread starter U Picciriddu
- Start date
we might be joking but stuff like this was unimaginable a few years ago:
https://www.independent.co.uk/tech/ai-ceo-artificial-intelligence-b2302091.html
i wonder which team will actually employ an ai as a coach first. i have no doubt that it will happen. unless federations ban it soon enough, that is.
https://www.independent.co.uk/tech/ai-ceo-artificial-intelligence-b2302091.html
i wonder which team will actually employ an ai as a coach first. i have no doubt that it will happen. unless federations ban it soon enough, that is.
Im cto of a venture backed company. I have 100+ developers in my company.
With the current LLM architecture (transformer), researchers has yet to see the ceiling, its just all about training with more data and more hardware. Researchers predicted we'll run out of quality data to train by 2026, so we are going to rely on synthetic data more
Google, amazon, facebook all have the money to keep advancing AI and openAI will get whatever money they needed too from investors.
Im just gonna pick software engineering cause thats the subject i understand the most, i believe we are not faraway from disrupting it. Devs are now already using assistant in their daily work, then checkout Devin from cognition labs - an AI coder, pretty mind blowing and it will only get better from here.
Heres the progression i foresee:
- Devs using AI to be more productive (now is already starting to become a norm, every devs in my company using github autopilot)
- Something like Devin allows innovative companies to start shifting their Dev's job to be more of architect and code reviewer.. this itself will create more data to train and improve improve the model. Not gonna lie, as CTO im excited to start incorporating something like this in my company
From the toolings perspectives, more and more toolings and platforms are going to be built in adjacent to AI advancements that will make solving problems even more trivial for AI to solve.
I see in less 5y companies dont need as many devs anymore compared to today and in less than 10y companies will only need software architects to build software solutions.
Maybe even sooner
- - - Updated - - -
Honestly right now is the gold rush
Just like dotcom gold rush in late 90s
Just like facebook apps gold rush in mid 2000s
Just like ios/android gold rush in late 2000s
Just like dApps gold rush (ICOs) not so long ago
You can make silly wrapper solution to solve niche problems using chatgpt or open source LLMs
I know some AI gf startup last year are making 100k/month.. and honestly its not hard to make at all
A friend of mine is paying 40/Month for silly air tratfic control AI cause hes a big flight sim nerd. Again, not hard to make.
Im currently thinking of building something as side project too.
I had dApp side project on ethereum last 2021 dApps / nft / defi gold rush making 500k usd in half a year (5 cofounders, we split evenly)
With the current LLM architecture (transformer), researchers has yet to see the ceiling, its just all about training with more data and more hardware. Researchers predicted we'll run out of quality data to train by 2026, so we are going to rely on synthetic data more
Google, amazon, facebook all have the money to keep advancing AI and openAI will get whatever money they needed too from investors.
Im just gonna pick software engineering cause thats the subject i understand the most, i believe we are not faraway from disrupting it. Devs are now already using assistant in their daily work, then checkout Devin from cognition labs - an AI coder, pretty mind blowing and it will only get better from here.
Heres the progression i foresee:
- Devs using AI to be more productive (now is already starting to become a norm, every devs in my company using github autopilot)
- Something like Devin allows innovative companies to start shifting their Dev's job to be more of architect and code reviewer.. this itself will create more data to train and improve improve the model. Not gonna lie, as CTO im excited to start incorporating something like this in my company
From the toolings perspectives, more and more toolings and platforms are going to be built in adjacent to AI advancements that will make solving problems even more trivial for AI to solve.
I see in less 5y companies dont need as many devs anymore compared to today and in less than 10y companies will only need software architects to build software solutions.
Maybe even sooner
- - - Updated - - -
Honestly right now is the gold rush
Just like dotcom gold rush in late 90s
Just like facebook apps gold rush in mid 2000s
Just like ios/android gold rush in late 2000s
Just like dApps gold rush (ICOs) not so long ago
You can make silly wrapper solution to solve niche problems using chatgpt or open source LLMs
I know some AI gf startup last year are making 100k/month.. and honestly its not hard to make at all
A friend of mine is paying 40/Month for silly air tratfic control AI cause hes a big flight sim nerd. Again, not hard to make.
Im currently thinking of building something as side project too.
I had dApp side project on ethereum last 2021 dApps / nft / defi gold rush making 500k usd in half a year (5 cofounders, we split evenly)
Last edited:
AI could easily automate many jobs ... until we bullshit them back into existence. I strongly recommend the 2018 book "Bullshit Jobs" by the late anthropologist David Graeber. The prevalence of "bullshit jobs" is also a reflection of broader economic and social structures. In most cases, jobs are created or maintained not because they are necessary for production or service, but to sustain employment levels or for bureaucratic reasons.
The same question was posed at the onset of every perceived technological revolution - from electrification and the assembly line to the Xerox and fax machine. Bullshit jobs proliferate because: humans.
In my opinion, LLMs like ChatGPT are just another tool at our disposal. Tools will always require some domain knowledge.
"Prompt Engineering" is very interesting. It's like the old story about someone charging $1,000 to fix a machine. They did it in 10 seconds and the client complained. The consultant said “I get paid for knowing how to fix it”. The same might be true for knowing how to prompt AI.
The same question was posed at the onset of every perceived technological revolution - from electrification and the assembly line to the Xerox and fax machine. Bullshit jobs proliferate because: humans.
In my opinion, LLMs like ChatGPT are just another tool at our disposal. Tools will always require some domain knowledge.
"Prompt Engineering" is very interesting. It's like the old story about someone charging $1,000 to fix a machine. They did it in 10 seconds and the client complained. The consultant said “I get paid for knowing how to fix it”. The same might be true for knowing how to prompt AI.
I’m not convinced enough yet. But I’m old enough to live through multiple AI winters before.
AI still basically just interpolates or extrapolates on a curve with dependent data. That’s it. When you make it sound like that, is hardly the sort of thing that would make you crap in your pants about the potential for the future. But there are a lot of low-hanging fruit and good use cases.
Gen AI, in specific, addresses some new and interesting problems. But as said here, the economics suck for most use cases. I.e., querying a compiled knowledge of the universe model to answer your 9th grade Algebra homework is a gross waste of resources relative to the results. Sure, you can fish with dynamite. But would you really want to make that your business model?
Generative AI is the hot topic, and everything is being thrown at it. AI has so much more to offer though. For example, causal AI with counterfactual modes sound way, way more impactful. But that’s not the It Girl right now.
Also, fundamentally I’m skeptical about digital-based AI systems because all of life is actually quantum, not binary. We are not 1s and 0s.
AI still basically just interpolates or extrapolates on a curve with dependent data. That’s it. When you make it sound like that, is hardly the sort of thing that would make you crap in your pants about the potential for the future. But there are a lot of low-hanging fruit and good use cases.
Gen AI, in specific, addresses some new and interesting problems. But as said here, the economics suck for most use cases. I.e., querying a compiled knowledge of the universe model to answer your 9th grade Algebra homework is a gross waste of resources relative to the results. Sure, you can fish with dynamite. But would you really want to make that your business model?
Generative AI is the hot topic, and everything is being thrown at it. AI has so much more to offer though. For example, causal AI with counterfactual modes sound way, way more impactful. But that’s not the It Girl right now.
Also, fundamentally I’m skeptical about digital-based AI systems because all of life is actually quantum, not binary. We are not 1s and 0s.
But while some of the underlying dimensions are exponential, the outward impacts are compressed on the higher end of an S-curve. Just think of giant particle accelerators: diminishing returns.
When your LLM indexes 100 billion documents, how much improvement do you get from 100 trillion? It’s not 1000x. It’s probably not even 5x.
Even to get to 100 trillion, your source quality has to go down so now you’re “contaminating” good data with Twitter tweets and Tuz posts.
So now when you query your LLM to produce a research paper on football transfer strategies in the age of Arab oil money, you get references to Allegri throaters. Garbage in, garbage out.
Which is why you now see AI execs freaking out that they feel they are soon to run out of source content. They say they can escape that problem through machine generated content. Which means you end up feeding the cow decomposed dead cow, and we know how that turns out. It’s what the AI folks call “model collapse”.
When your LLM indexes 100 billion documents, how much improvement do you get from 100 trillion? It’s not 1000x. It’s probably not even 5x.
Even to get to 100 trillion, your source quality has to go down so now you’re “contaminating” good data with Twitter tweets and Tuz posts.
So now when you query your LLM to produce a research paper on football transfer strategies in the age of Arab oil money, you get references to Allegri throaters. Garbage in, garbage out.
Which is why you now see AI execs freaking out that they feel they are soon to run out of source content. They say they can escape that problem through machine generated content. Which means you end up feeding the cow decomposed dead cow, and we know how that turns out. It’s what the AI folks call “model collapse”.
Oh man... what am I in for?
Now the whole AI orgy over generative AI is a bit of an Inter fake orgasm fest in my book. Some applications yes. But too many people banking their entire career futures on it.
That said, what makes me curious is more the periphery than the overexposed core. Doomers aside, the question of philosophy and ethics as it intersects with AI is a curious one to me ... something I wish we had paused before diving head-first into toxic social media. So recently I applied for a global AI governance and ethics program and was recently accepted to the cohort. Class starts next week.
Now the fellow new students are rolling in on the Slack #intros channel. I'm already getting nervous about all the Gen Z dweebs professing their SBF-like Effective Altruism (EA) fanboiness ... it's become the new freshman in college who reads Ayn Rand's Fountainhead for the first time.
And they invariably look like former Young Republican gestapo boys like this dude (real photo of a real fellow student):
TL;DR: our world is f*&ked
Now the whole AI orgy over generative AI is a bit of an Inter fake orgasm fest in my book. Some applications yes. But too many people banking their entire career futures on it.
That said, what makes me curious is more the periphery than the overexposed core. Doomers aside, the question of philosophy and ethics as it intersects with AI is a curious one to me ... something I wish we had paused before diving head-first into toxic social media. So recently I applied for a global AI governance and ethics program and was recently accepted to the cohort. Class starts next week.
Now the fellow new students are rolling in on the Slack #intros channel. I'm already getting nervous about all the Gen Z dweebs professing their SBF-like Effective Altruism (EA) fanboiness ... it's become the new freshman in college who reads Ayn Rand's Fountainhead for the first time.
And they invariably look like former Young Republican gestapo boys like this dude (real photo of a real fellow student):
TL;DR: our world is f*&ked
@swag I recently read an article where the writer did some research and found something like 155 scholarly articles with the phrase “thats as far back as my knowledge goes” or something similar. It meant they used AI to generate the articles and the journal was riddled with them. These are only the ones without spell checking.
Which begs the question, will these then be fed back to the AI as data? So AI will regenerate data based on data it created itself. Resulting in a shitshow of nonsense.
Which begs the question, will these then be fed back to the AI as data? So AI will regenerate data based on data it created itself. Resulting in a shitshow of nonsense.
