No, that’s google. A leaked memo indicates that they deliberately made search worse so people would have to search twice and see twice the number of ads
AI companies lose money every time someone uses it. Even those that charge per-token
Pretty amazing what PiHole reveals. My work laptop constantly chatters with sites all over the world, it's the biggest source of DNS lookup in my home network while it's on, and it's competing with six people browsing the web for their various purposes...
They're burning billions because they're trying to rush ahead of the competition in capabilities. No matter how good LLMs get, that is not their goal. They're trying to reach AGI and there are no second places in that race.
Thinking that LLMs will ever become AGI is fucking hysterical, and that is what the shills keep saying is going to happen. They are trying to turn lead into gold using a stove and a skillet.
No money they have dumped into LLMs is going to contribute to something that could achieve AGI. They are running the wrong race.
Then I don't see how your response in any way relates to what I said.
Whether you think they'll ever get there or not is completely irrelevant. That is still what they're after and the reason for the massive upfront investments.
There is no AGI. Whatever they reach, there will always be a next level.
In the 1980s if you said you could make a computer play Go better than the best human masters, there are those who would have said "that would be true artificial intelligence." Then it happened, and the world moved on.
There was a time when a computer understanding speech, translating languages, would have been considered true artificial intelligence, then we got there, and the world moved on.
LLMs have solved a mathematical question left open with prize money for decades unsolved by humans (just one, really, that I know of, so far...), but that's not AGI yet...
Many forms of "the Turing Test" are being passed by LLMs tested against the majority of the general population now, but apparently that's not AGI yet...
AGI will continue to be a moving goalpost, as it should be. It's not a finish line, it's a journey. Even when automated systems are building themselves from raw material inputs, designing and building their own infrastructure, power plants, communications, and continuously improving their own designs, there will be those who still design new tests for "AGI" that they don't pass, yet.
AI and AGI are not synonymous terms. We've had AI since 1956. General intelligence means human-level intelligence. When an AI system can do any task as well as or better than humans can, it's by definition generally intelligent.
We're not changing the definitions. People thought that chess is so hard that once an AI system can play chess it has to be as intelligent as humans. That just turned out to be a false assumption. A system can be superhuman at playing chess but that ability doesn't need to translate to any other field.
General intelligence means human-level intelligence.
Applicable quote from a Yosemite park ranger: "There is considerable overlap between the intelligence of the smartest bears and the dumbest tourists."
We’re not changing the definitions. People thought that chess is so hard... That just turned out to be a false assumption
Sounds like changing definitions to me.
A system can be superhuman at playing chess but that ability doesn’t need to translate to any other field.
AlphaZero has been superhuman at playing chess, Go, and basically any game with perfect information and fixed rules since early 2018. That's translatable to other fields - it's not as strong in some fields as it is in playing Go, but it's still translatable ability...
But it's not. It's a hypothesis that was proven incorrect by running the experiment.
That’s translatable to other fields
I didn't say it doesn't ever translate to other fields. I said that it doesn't need to. That's the difference between narrow and general intelligence. You can be superhuman at X but have zero clue how to do Y. It's still intelligence but it's not general or "human level"
I think they know AGI is far off, and are setting a more medium term goal. I think they are just trying to corner the new LLM market that might emerge. Even if the bubble pops and 3/4 go bust, they hope to be the one that survives and gets to be the quasi-monopoly in that market for the next decade.
Nobody "knows" that AGI is far off just like nobody knows that it's near either. That is not an established fact. Those are both just popular narratives in their respective camps. We could get there tomorrow or it could take usanother 400 years. Both are plausible outcomes.
That is just utter bullshit. Hallucinations are a by-product of how LLMs work under the hood, not an intentional design choice. An AI that doesn't make mistakes would be orders of magnitude more profitable.
The prevalence of hallucination in LLMs is a design choice. It is a result of raising the 'temperature' which is just fancy speak for randomization so it doesn't spit out the same text for the same question over and over to make it look like it has nuance and whatever.
If it was consistent they would be able to reduce incorrect results, but they want it to look like a human response.
It's not just "looking like a human response" it's also functioning like a human response. The randomness of results enables iterative soltions that make forward progress instead of getting stuck.
There are a vast set of problems which don't have a single perfect "correct" answer where all others are wrong, there are just collections of "answers" which - when taken as a set - work together to form a working solution. You may have 100 questions to answer, and how you answer the first 10 will affect what does or does not work for the next 10, and the next, down the line, and you may find when you get to the last set of 10 that you can't get to the end solution unless you revise some of the answers that you previously gave - answers that looked resonable until you built the next 80% of the product...
Life isn't school - there aren't 10 question quizzes with pick one of 4 multiple choice possible answers where you can get a perfect score just by answering each question correctly one at a time. Real-life school is being in charge of class assignments for 1000 students, chosing which 25 students go in each room with each teacher. What classes do they get, what combinations of students should be kept together, kept apart, grouped with which teachers... they aren't impossible problems, but they are impossible to optimize for all possible considerations. Tradeoffs have to be made.
They're getting a little better about that as time goes on, but yeah, last year the time blindless was a major handicap at times.
On questions like geometrically constrained requirements, they're pretty good at telling you when a problem is overconstrained such that there is no answer, but... in the fuzzier world of underspecified questions, they'll stretch pretty far to make up an answer. In the world of computer programming, sometimes that's a brilliant move - they "make up" some code, compile it, test it, and it works - it's actually a functional solution.
The other day I challenged Gemini to find a person that I had a vague description of, Gemini went out and made up a name, job title, vague description of their publication history. When I pressed for actual evidence, its answers were evasive, and when I finally cornered it with a demand for anything concrete proving this person actually exists it came clean with "I hallucinated that."
Mistakes are part of the human process... an automaton which produces only one solution for a problem is easily stuck, trapped, dead-ended. Building imperfect solution candidates and improving them until they are acceptable is how humans have designed things since forever. There are no perfect answers to the questions that matter.
I partially agree with the idea that AI is the same as industrial revolution. Yes, AI is just as revolutionary as steam engine.
Except we are currently at the stage steam engine was in ancient Greece and Roman Empire - a curiosity too expensive to use for value it provides and too crude to provide anything better.
Luckily ancient times didn't have billionaires playing speculative gambling on economy by trying to push any new thing as something revolutionizing right now.
Let the AI flop now, wait a few centuries and it will return better.
AI will cause next productivity revolution when demand for things that AI can make will heavily outweigh supply created by human workforce. This is not the case now. We automated work of artists, people we generally joke about being unneeded beggars.
Industrial revolution happened because demand for goods was so big, people would rather buy mediocre factory-made things than artisan-made high quality goods. Common folk currently do not have enough money to spend on frivolous things AI makes.
The thing about money - the only thing money is actually good for is: manipulating the behavior of humans.
You can't eat it, but you can get people to give you food in exchange for it. Money doesn't grow food, but it does get people to do the work of growing and harvesting and shipping and preparing food...
Money doesn't build your house, but it does get people to build a house for you. It doesn't make electricity, but it does get people to build and operate electrical generation systems...
So, what are Billionaires? They're actually just power-centers of people manipulation by money.
This is inaccurate. What's currently happening is that the AI companies are throwing away money because they're riding the bubble.
What they pretend they'll be able to sell is a replacement employee. Of course we've seen that in most fields, AI is not nearly good enough to replace us, and probably in most fields it never will be.
What you're pointing out is one way to generate more consumption of AI, and that would be monetizable if AI were actually making money on queries. But right now it's losing money on those queries.
Also, when you get used to relying on ai, you lose the practice and forget part fo your knowledge and skills. So, if you try to stop using ai, you will first have a steep decrease in productivity as you have to resharp your skills and remember a lot of stuff, and that creates a barrier preventing people from going out
In the 1980s I worked at a factory where the joke was "why are coffee breaks only 15 minutes long?" "Because when we give 'em 30 minutes for lunch we have to retrain them before they start work again after lunch."
The "better models" have been interesting to watch progress over the past year. I'd say the free to use models today are better than the best that were available a year ago. The ones with bigger context windows use more resources, and sometimes can give better results, often not. In LLMs, management of what is, and is not, in the context window seems to be the key to the kinds of results you get, and it feels like they have been "learning" to self-manage their context windows quite a bit better over the past 12 months.
I agree. Over time I have learned to be a lot more careful with the context window and periodically start over to keep it small. This was one of the reasons I left the free ChatGPT, it seemed to have a very small context window and was not graceful at all about going outside it. Gemini free tier was a lot more graceful about this. I think the advantage of the paid tiers is simply that they will try to manage for longer and report to you how big your context window has gotten. So you have more time and you know when to start thinking about starting from scratch again.
I haven't tried lately, several months ago I tried asking the chatbots directly: What's the size of your context window. Gemini answered straight out: "32,767 tokens, and that's not as good for developing complex software as a larger context window like Claude Sonnet's 200,000 tokens."
Gemini is 1 million now. But you should probably stop before then. And yes, it's surprisingly honest about whether it's the right model for your needs. It's recommended me to go with Claude for some of my projects.
Back when Sonnet was 200K and Opus was 1M, there were a lot of complex programming projects where I actually got better overall results out of Sonnet... but, go back to the 3.x days and Sonnet got stuck in debug loops fairly often where Opus would break out of the loop and find a working solution more often.
Funny that, I've learned how to spell that word in four different languages. But mistakes tend to happen when typing quickly on a phone. Your concern and attempt at deflection is noted.
I use AI for first drafts of work frequently. I'm also in the process of building a chief of staff agent, which it pretty cool. I pay $30/month. Not bad.
It's not that. It's the idea that by being right just a bit more times than wrong, statistically this would bring out good results if you try it enough times. If that doesn't happen, it's really your fault 😀.
This is the power of specifications. With a specification, the LLM can create a test to ensure what it creates meets the specification. Then the agent can iterate on solution after solution until its proposed solution meets all your specifications. Then you get to discover how many holes you left open in your specifications.
Another thing AI agents based on LLMs aren't half bad at: reviewing specifications for completeness and readiness for implementation. They'll go ahead and fill in the blanks for you if you ask them to, but if you ask them instead to point out the holes then you can decide what should be done instead of "yeah, just make mine like all the other ones you have seen in your training."
60 Comments
SaraTonin@lemmy.world · 26 pts · 38d
No, that’s google. A leaked memo indicates that they deliberately made search worse so people would have to search twice and see twice the number of ads
AI companies lose money every time someone uses it. Even those that charge per-token
thermal_shock@lemmy.world · 7 pts · 38d
I refuse to use a computer or browser without ublock origin. No fucking way I'm accepting that poisonous shit they call ads.
MangoCats@feddit.it · -4 pts · 38d
https://www.reddit.com/r/technology/comments/1u4q5e0/the_end_of_ublock_origin_in_chrome_is_now_weeks/
thermal_shock@lemmy.world · 11 pts · 38d
I've been on Firefox for the last year where it does work. I don't use chrome anymore.
I also host a pihole at home blocking stuff too.
Stands works well though, saw a client using it recently
https://chromewebstore.google.com/detail/ad-blocker-stands-adblock/lgblnfidahcdcjddiepkckcfdhpknnjh?hl=en-US&pli=1
MangoCats@feddit.it · 2 pts · 38d
Pretty amazing what PiHole reveals. My work laptop constantly chatters with sites all over the world, it's the biggest source of DNS lookup in my home network while it's on, and it's competing with six people browsing the web for their various purposes...
thermal_shock@lemmy.world · 1 pts · 38d
Oh yeah, very eye-opening the amount of data it's sending/receiving. Looking into gateways with VPN built in.
ViscloReader@lemmy.world · 2 pts · 38d
Omg please link
TheDankMemegician36@thelemmy.club · 20 pts · 39d
This is the entire business model for most if not all of tech really
snooggums@piefed.world · 15 pts · 39d
If it worked they could be profitable already instead of continuing to burn billions of investment dollars.
Iconoclast@feddit.uk · 1 pts · 39d
They're burning billions because they're trying to rush ahead of the competition in capabilities. No matter how good LLMs get, that is not their goal. They're trying to reach AGI and there are no second places in that race.
snooggums@piefed.world · 25 pts · 38d
Thinking that LLMs will ever become AGI is fucking hysterical, and that is what the shills keep saying is going to happen. They are trying to turn lead into gold using a stove and a skillet.
No money they have dumped into LLMs is going to contribute to something that could achieve AGI. They are running the wrong race.
Iconoclast@feddit.uk · -8 pts · 38d
Mind quoting the part where I claimed LLMs will become AGI?
snooggums@piefed.world · 8 pts · 38d
Altman and the other shills claim that.
Are you an AI shill?
If not, then I didn't say you claimed that.
Iconoclast@feddit.uk · -7 pts · 38d
Then I don't see how your response in any way relates to what I said.
Whether you think they'll ever get there or not is completely irrelevant. That is still what they're after and the reason for the massive upfront investments.
MangoCats@feddit.it · 3 pts · 38d
There is no AGI. Whatever they reach, there will always be a next level.
In the 1980s if you said you could make a computer play Go better than the best human masters, there are those who would have said "that would be true artificial intelligence." Then it happened, and the world moved on.
There was a time when a computer understanding speech, translating languages, would have been considered true artificial intelligence, then we got there, and the world moved on.
LLMs have solved a mathematical question left open with prize money for decades unsolved by humans (just one, really, that I know of, so far...), but that's not AGI yet...
Many forms of "the Turing Test" are being passed by LLMs tested against the majority of the general population now, but apparently that's not AGI yet...
AGI will continue to be a moving goalpost, as it should be. It's not a finish line, it's a journey. Even when automated systems are building themselves from raw material inputs, designing and building their own infrastructure, power plants, communications, and continuously improving their own designs, there will be those who still design new tests for "AGI" that they don't pass, yet.
Iconoclast@feddit.uk · 4 pts · 38d
AI and AGI are not synonymous terms. We've had AI since 1956. General intelligence means human-level intelligence. When an AI system can do any task as well as or better than humans can, it's by definition generally intelligent.
We're not changing the definitions. People thought that chess is so hard that once an AI system can play chess it has to be as intelligent as humans. That just turned out to be a false assumption. A system can be superhuman at playing chess but that ability doesn't need to translate to any other field.
MangoCats@feddit.it · 0 pts · 37d
Applicable quote from a Yosemite park ranger: "There is considerable overlap between the intelligence of the smartest bears and the dumbest tourists."
Sounds like changing definitions to me.
AlphaZero has been superhuman at playing chess, Go, and basically any game with perfect information and fixed rules since early 2018. That's translatable to other fields - it's not as strong in some fields as it is in playing Go, but it's still translatable ability...
Iconoclast@feddit.uk · 1 pts · 37d
But it's not. It's a hypothesis that was proven incorrect by running the experiment.
I didn't say it doesn't ever translate to other fields. I said that it doesn't need to. That's the difference between narrow and general intelligence. You can be superhuman at X but have zero clue how to do Y. It's still intelligence but it's not general or "human level"
Kazumara@discuss.tchncs.de · 2 pts · 38d
I think they know AGI is far off, and are setting a more medium term goal. I think they are just trying to corner the new LLM market that might emerge. Even if the bubble pops and 3/4 go bust, they hope to be the one that survives and gets to be the quasi-monopoly in that market for the next decade.
Iconoclast@feddit.uk · -2 pts · 38d
Nobody "knows" that AGI is far off just like nobody knows that it's near either. That is not an established fact. Those are both just popular narratives in their respective camps. We could get there tomorrow or it could take usanother 400 years. Both are plausible outcomes.
Kazumara@discuss.tchncs.de · 2 pts · 37d
Sure just like we could discover the Greek gods exist tomorrow.
The point is that there is no reason to plan towards AGI for those CEOs.
Iconoclast@feddit.uk · 0 pts · 37d
Ridicule is evasion, not counter.
Kazumara@discuss.tchncs.de · 2 pts · 37d
That's okay with me. This discussion hasn't been fruitful since you latched on to the term "know" instead of engaging with the content.
Iconoclast@feddit.uk · 12 pts · 39d
That is just utter bullshit. Hallucinations are a by-product of how LLMs work under the hood, not an intentional design choice. An AI that doesn't make mistakes would be orders of magnitude more profitable.
snooggums@piefed.world · 1 pts · 38d
The prevalence of hallucination in LLMs is a design choice. It is a result of raising the 'temperature' which is just fancy speak for randomization so it doesn't spit out the same text for the same question over and over to make it look like it has nuance and whatever.
If it was consistent they would be able to reduce incorrect results, but they want it to look like a human response.
lIlIlIlIlIlIl@lemmy.world · 1 pts · 38d
Can you provide sources to “they want it to look like a human response?”
I have not read about that before.
snooggums@piefed.world · 0 pts · 38d
The whole idea of LLMs is to replicate human language, as in have LLM output replicate language spoken by humans.
Here's something about improving it: Enhancing Human-Like Responses in Large Language Models
Here's a big thing about temperature: https://www.ibm.com/think/topics/llm-temperature
MangoCats@feddit.it · 0 pts · 38d
It's not just "looking like a human response" it's also functioning like a human response. The randomness of results enables iterative soltions that make forward progress instead of getting stuck.
There are a vast set of problems which don't have a single perfect "correct" answer where all others are wrong, there are just collections of "answers" which - when taken as a set - work together to form a working solution. You may have 100 questions to answer, and how you answer the first 10 will affect what does or does not work for the next 10, and the next, down the line, and you may find when you get to the last set of 10 that you can't get to the end solution unless you revise some of the answers that you previously gave - answers that looked resonable until you built the next 80% of the product...
Life isn't school - there aren't 10 question quizzes with pick one of 4 multiple choice possible answers where you can get a perfect score just by answering each question correctly one at a time. Real-life school is being in charge of class assignments for 1000 students, chosing which 25 students go in each room with each teacher. What classes do they get, what combinations of students should be kept together, kept apart, grouped with which teachers... they aren't impossible problems, but they are impossible to optimize for all possible considerations. Tradeoffs have to be made.
snooggums@piefed.world · 2 pts · 38d
Tradeoffs like not understanding time and always giving an answer even when there isn't one.
MangoCats@feddit.it · 0 pts · 37d
They're getting a little better about that as time goes on, but yeah, last year the time blindless was a major handicap at times.
On questions like geometrically constrained requirements, they're pretty good at telling you when a problem is overconstrained such that there is no answer, but... in the fuzzier world of underspecified questions, they'll stretch pretty far to make up an answer. In the world of computer programming, sometimes that's a brilliant move - they "make up" some code, compile it, test it, and it works - it's actually a functional solution.
The other day I challenged Gemini to find a person that I had a vague description of, Gemini went out and made up a name, job title, vague description of their publication history. When I pressed for actual evidence, its answers were evasive, and when I finally cornered it with a demand for anything concrete proving this person actually exists it came clean with "I hallucinated that."
MangoCats@feddit.it · 1 pts · 38d
Mistakes are part of the human process... an automaton which produces only one solution for a problem is easily stuck, trapped, dead-ended. Building imperfect solution candidates and improving them until they are acceptable is how humans have designed things since forever. There are no perfect answers to the questions that matter.
Yaky@slrpnk.net · 11 pts · 37d
from Don't Look Up.
HrabiaVulpes@europe.pub · 7 pts · 38d
I partially agree with the idea that AI is the same as industrial revolution. Yes, AI is just as revolutionary as steam engine.
Except we are currently at the stage steam engine was in ancient Greece and Roman Empire - a curiosity too expensive to use for value it provides and too crude to provide anything better.
Luckily ancient times didn't have billionaires playing speculative gambling on economy by trying to push any new thing as something revolutionizing right now.
Let the AI flop now, wait a few centuries and it will return better.
irish_link@lemmy.world · 2 pts · 37d
Better in terms of using less resources would be great.
HrabiaVulpes@europe.pub · 3 pts · 37d
AI will cause next productivity revolution when demand for things that AI can make will heavily outweigh supply created by human workforce. This is not the case now. We automated work of artists, people we generally joke about being unneeded beggars.
Industrial revolution happened because demand for goods was so big, people would rather buy mediocre factory-made things than artisan-made high quality goods. Common folk currently do not have enough money to spend on frivolous things AI makes.
EyIchFragDochNur@lemmy.world · 6 pts · 39d
And when we're done fixing, we're unnecessary and have time to eat the rich \o/
One can dream .. why not replace billionaires with AI too?
MangoCats@feddit.it · 1 pts · 38d
The thing about money - the only thing money is actually good for is: manipulating the behavior of humans.
You can't eat it, but you can get people to give you food in exchange for it. Money doesn't grow food, but it does get people to do the work of growing and harvesting and shipping and preparing food...
Money doesn't build your house, but it does get people to build a house for you. It doesn't make electricity, but it does get people to build and operate electrical generation systems...
So, what are Billionaires? They're actually just power-centers of people manipulation by money.
fodor@lemmy.zip · 6 pts · 38d
This is inaccurate. What's currently happening is that the AI companies are throwing away money because they're riding the bubble.
What they pretend they'll be able to sell is a replacement employee. Of course we've seen that in most fields, AI is not nearly good enough to replace us, and probably in most fields it never will be.
What you're pointing out is one way to generate more consumption of AI, and that would be monetizable if AI were actually making money on queries. But right now it's losing money on those queries.
morto@piefed.social · 6 pts · 38d
Also, when you get used to relying on ai, you lose the practice and forget part fo your knowledge and skills. So, if you try to stop using ai, you will first have a steep decrease in productivity as you have to resharp your skills and remember a lot of stuff, and that creates a barrier preventing people from going out
MangoCats@feddit.it · 0 pts · 38d
Jesus, people, AI has barely been usable for a year - do you all forget your years of study and practice if the coffee breaks are too long, too?
morto@piefed.social · 4 pts · 38d
Many people will have already forgotten what they studied in the first semesters when they finish the course lol
WIthout constant practice, most people's skills get rusty in a faster rate than we can imagine
MangoCats@feddit.it · 4 pts · 38d
In the 1980s I worked at a factory where the joke was "why are coffee breaks only 15 minutes long?" "Because when we give 'em 30 minutes for lunch we have to retrain them before they start work again after lunch."
realitista@lemmus.org · 5 pts · 39d
Unless you pay more for the better model, then it makes slightly less mistakes.
MangoCats@feddit.it · 1 pts · 38d
The "better models" have been interesting to watch progress over the past year. I'd say the free to use models today are better than the best that were available a year ago. The ones with bigger context windows use more resources, and sometimes can give better results, often not. In LLMs, management of what is, and is not, in the context window seems to be the key to the kinds of results you get, and it feels like they have been "learning" to self-manage their context windows quite a bit better over the past 12 months.
realitista@lemmus.org · 1 pts · 38d
I agree. Over time I have learned to be a lot more careful with the context window and periodically start over to keep it small. This was one of the reasons I left the free ChatGPT, it seemed to have a very small context window and was not graceful at all about going outside it. Gemini free tier was a lot more graceful about this. I think the advantage of the paid tiers is simply that they will try to manage for longer and report to you how big your context window has gotten. So you have more time and you know when to start thinking about starting from scratch again.
MangoCats@feddit.it · 1 pts · 37d
I haven't tried lately, several months ago I tried asking the chatbots directly: What's the size of your context window. Gemini answered straight out: "32,767 tokens, and that's not as good for developing complex software as a larger context window like Claude Sonnet's 200,000 tokens."
realitista@lemmus.org · 1 pts · 37d
Gemini is 1 million now. But you should probably stop before then. And yes, it's surprisingly honest about whether it's the right model for your needs. It's recommended me to go with Claude for some of my projects.
MangoCats@feddit.it · 1 pts · 37d
Back when Sonnet was 200K and Opus was 1M, there were a lot of complex programming projects where I actually got better overall results out of Sonnet... but, go back to the 3.x days and Sonnet got stuck in debug loops fairly often where Opus would break out of the loop and find a working solution more often.
FriendOfDeSoto@startrek.website · 5 pts · 39d
and then fix the new mistakes it made while trying to fix the old ones and
Atomic@sh.itjust.works · 3 pts · 37d
Not to be that guy, but that's some weak ass strawman bullshit.
Surely you can think of a more solid point of critisism than whatever that was.
jaykrown@lemmy.world · 0 pts · 37d
Surely you can learn to spell "criticism" correctly.
Atomic@sh.itjust.works · 5 pts · 37d
Funny that, I've learned how to spell that word in four different languages. But mistakes tend to happen when typing quickly on a phone. Your concern and attempt at deflection is noted.
SystemQ@lemmy.world · 2 pts · 37d
Kolanaki@pawb.social · 2 pts · 38d
I don't need a machine to make mistakes. 😤
dream_weasel@sh.itjust.works · 1 pts · 38d
"Drug companies sell drugs to make you more sick!" vibes.
lIlIlIlIlIlIl@lemmy.world · 0 pts · 39d
I use local AI models to improve my process. I pay zero, and they do a pretty good job of taking the grunt work out of my tasks
sunbrrnslapper@lemmy.world · 0 pts · 38d
I use AI for first drafts of work frequently. I'm also in the process of building a chief of staff agent, which it pretty cool. I pay $30/month. Not bad.
MangoCats@feddit.it · 2 pts · 38d
I use AI to review proposed final drafts a lot. It finds all kinds of nuance (and problems) in minutes that would take me hours to find without it.
PixelatedSaturn@lemmy.world · -2 pts · 39d
It's not that. It's the idea that by being right just a bit more times than wrong, statistically this would bring out good results if you try it enough times. If that doesn't happen, it's really your fault 😀.
MangoCats@feddit.it · 1 pts · 38d
This is the power of specifications. With a specification, the LLM can create a test to ensure what it creates meets the specification. Then the agent can iterate on solution after solution until its proposed solution meets all your specifications. Then you get to discover how many holes you left open in your specifications.
Another thing AI agents based on LLMs aren't half bad at: reviewing specifications for completeness and readiness for implementation. They'll go ahead and fill in the blanks for you if you ask them to, but if you ask them instead to point out the holes then you can decide what should be done instead of "yeah, just make mine like all the other ones you have seen in your training."