Can you elaborate?
Is it because the models are inherently not deterministic? But then, will it not get better with time? I mean arent we just at the beginning of the research here?
It is because English is an ambiguous input unlike a high level programming language. The output cannot help but be wrong sometimes when the input is ambiguous. See my comment here https://news.ycombinator.com/item?id=49491797
Determinism isn't the issue; it's that the prompt does not contain enough information to know the correct way to do the thing, even if it did the thing the same way with the same inputs every time.
> Is it because the models are inherently not deterministic
This is the main reason (and yes, many modern compilers and CPUs carry some non-determinism which actually quite well explained and specified).
> But then, will it not get better with time?
No, it won't. Bacuse that is that is the actual literal limitation/feature of LLMs.
> I mean arent we just at the beginning of the research here?
In general? Yes. With LLMs? We can reasonably say that they will never be deterministic.
There is a way to get a non-determenistic output: first question on a temp 0 local model on a completely new session will give you the same answer. The second answer in that same session will already be different on every session (even if it's the same question).
Honestly big noobquestion: isn't math just very very nested patternmatching based on a few foundational operators?
ive always felt, that im bad at math, cause i forget all the rules, but seeing solutions (and knowing the used pattern) always made "sense".
I always thought the hard math problems are so deeply nested or you have to remember trick xyz that people just didnt think about it yet..
The amount of mathematical structures and transformations you can apply (the possible rules) is effectively infinite. Simply remembering the rules might work at first, but you'll soon run into the combinatorial explosion: https://en.wikipedia.org/wiki/Combinatorial_explosion
You could go a step further, and simply say "well, ok, then the LLMs are merely doing some form of incremental/heuristic search!". Yes, but at that point you'd also be hard-pressed to claim that humans themselves are doing anything beyond that. You run out of naturalistic explanations.
I think this is a very subjective matter and depends on how negatively connoted someone's perception of the word 'manipulate' is. By your definition, I would consider 'studying/learning' also a form of manipulation.
For me its more his attitude that puts me off.
He might be intelligent, but his EQ doesnt seem that high.
The condescending way he references the "malcolm in the middle"-episode "hot dumb girl" couldve been just the explaination of the "1 dollar = 1 million dollar".
Its baffling to me how the US cannot handle their traffic laws.
How is there any doubt in running a red light? Why can they not let common sense handle it?
Just fine the car owner without question.
Car owners will think twice borrowing their cars.
Easy detection, less bureaucracy. And hopefully safer streets.
Read other comments. It is not so simple. Traffic lights often are configured wrongly, when yellow light is too short. So, violations are used to profit from cases when the driver could not have been stopped. In such system, it is better to be safe than sorry.
What google search alternative have you found?
Im trying out ecosia, duckduckgo and brave search, but i find their search results even worse, so in the second query i tend to bang to google..
Google Search is over. There may not be a free alternative, it they've lost the arms war between phone number incrementing ad pages, AI spew, and rank hackers.
You make it sound like a significant amount is going to Kreml but I assume the API cost for using Yandex from Kagi is neglectable and only a fraction of that goes to the Russian government. Isn't this more of a symbolic thing to request not cooperating with Russian companies?
I think that "it's better to know" only really holds up if the scope / context is also included. To put it in concrete terms, I'd amend your statement like this:
Kagi indirectly funds the Kremlin's regime by paying for Yandex API access.
Agree. Historically you would just not get any good results for a search and try Google, but these days it's more likely there just aren't any good results for your search period, regardless of engine. Funny enough that's when I've had better results asking chatgpt or similar because I'm typically after some sort of consensus or summary in those situations.
DDG is good enough that I've switched many year ago and never went back. Any time I use Google (!g) to repeat query (recently it's maybe a few times per year) it fails to show anything useful too, so I don't see any benefit to even check it lately.
Similar experience for me. I've been using DDG for years and while the quality has gone downhill somewhat I still rarely use !g because Google almost never has a useful result either if DDG strikes out.
Maybe have to pay for search? I am experimenting with paying Proton another $10/month for a paid lumo+ account. lumo+ is a private chat like ChatGPT that uses a strong Mistral model and also privacy-preserving web_search LLM tooling under the hood. For about a month I just use lumo+ with the web_search tool enabled. I may not do this forever, but for now I like just having one tool to use. Note: I still use gemini for technical work, but lumo+ for day to day chat and web search.
In the past I just use DuckDuckGo for most search, occasionally Google. That also worked well for me.
Honestly imo the driving license requirements and the respective fines for violations is too low. Rigid rules generally improve the traffic flow. But as soon as someone just doesnt care, the system breaks down.
Why forget about boxes and deterministic control and start thinking of error tolerance and recovery?
I know, that LLMs are statistical models, but can you not use patterns to enforce a deterministic outcome? (Single responsibility for each agent, retrying llm calls, rephrasing prompts, etc?)
Hopefully it can fill the void in Linux and Windows Systems.
And please without subscription. If it works fine id love to pay 50$+ like the Alfred Powerpack
reply