The AI of Google resolv a lot of computers problems

Part of the issue with AI is that it allows you to skip the stage where you know what it is that you're asking for. If you can't work out what you need, then you're not likely to identify if there are any issues with the proposed solution. I've also personally found that when I do try using AI that it basically never gives me anything that's as useful as a forum post written by humans that know what they're doing and often understand why it's the right answer.
 
The Intel wireless cards are limited to the 2.4G band as access point. The channels of the 5G band are all disabled by default and I have not find a way to enable the 5G channels who are not used by the army and the meteo. As client it work well.
 
Hi
It is a logic machine, nothing more; through our questions, we contribute to its learning process, help improve the model, and enable it to store a vast amount of data.
We must stop viewing these machines as thinking entities.
It feeds on a multi-level search of *our* knowledge.
It starts with observations and answers (whether accurate or incorrect) and "spits out" a synthesis.
This synthesis becomes part of the sum of our knowledge.
Often true, sometimes false.
 
I believe that some people have had success using wifibox, where you install a small alpine linux VM and run the card through that. Later Intel cards will work at 5G speeds. I have one headless Beelink with an AX200. Using https://www.speedtest.net/apps/cli, it has a download speed of 351.75 Mb/s from my provider (Spectrum and to save money we aren't on the fastest tier.) In contrast a wired machine (RTL811/8168/8211) is around the same, 390 Mbs last I tested.
 
According to the AI of google the developpers team of FreeBSD is focussed on the managed mode of the wireless cards for laptops. It can take years for them to enlarge the support for access point in 5G and 6G. Of course I have not question the developpers about this.
 
DanielBowman said:
AI, even the free one, is very helpful with prototyping.
It gives you basic syntax and options fast, even the general logic/schema. No need to sift through google searches (mediocre general reddit posts mostly) and copy/pasting code from stackoverflow like in the old days.
Then you can quickly test the proof of concept, for sometimes it reads some obscure post on a forum, only suggesting some feature but it "thinks" the feature is there and even hallucinates syntax for it.
For production I would recommend building it block by block, segment by segment. Actually, exactly with accordance with Unix philosophy.
Some coding harness (Pi, opencode] can probably manage this on it's own. Plus being documentation on the fly. But with the risk of losing perspective - crucial if buggy.

If I embark on some "bigger" project, I usually search for Video proof. Some guy showing it working on youtube. In my experience, there is a lot of brainiacs who suggest to you solutions which should work in theory from behind the desk but in reality you run in to million nuances, sometimes fatal i.e. manufacturer did not implement a feature. It should work but it just doesn't.
Click to expand...
This is exactly how I use any AI tool, and more importantly that the free versions have always been more than enough for me. The chatbot craze started around my sophomore year of college so I have a smaller sample size of learning syntax and discovering bugs before then, especially since I was not doing anything complex in high school haha. Although maybe this was the best time to be exposed to a “new”technology?!

I would say a chatbot makes information retrieval more specialized yet accessible. In a way, it lets people without the knowledge on how to search, search. The ability to ask follow up queries and probabilistic filtering(the two things that make an LLM) is why chatbots can be much more powerful than search engines, especially for the general population with general type questions.

For technical questions I’ve found that it is best practice to craft a query where you can describe each code block by inputs, outputs, and type of computation in the middle. Not only are you 80% of the way there by understanding what you are trying to do, but what’s left is the syntax itself. These queries normally result in anywhere from 5-25 LOC. Writing SQL code, scripting languages, finding and using libraries, runtime error debugging, and using OS system tools are all things I’ve found help with chatbots. I will say these bots are not great at HDL like VHDL or Verilog, really anything electrical. I’ve been happy in computer engineering with an emphasis on hardware!

I’ve met too many people with no knowledge, using the latest models of anthropic to generate heaps of code, connecting to multiple subscription based APIs, and ripping through half of their daily tokens in single prompts. All while acting like they built the system. No, unfortunately it is a template-esque site that uses expensive APIs for everything and would not only be outrageously expensive for any amount of users, but would not scale.

To top it off, those people look at me crazy when I say I use the free version!
Logicien said:
I only ask the AI for things that I cannot do by myself. The FreeBSD documentation is just a part of what you can read of FreeBSD on Internet. To say the true I often want to search for an immediate solution without headatche. I try to learn because the solution can take time with many tries by the AI. AI is the last resource when no other things work. They have beat humans in chess and can beat us too in some other fields.
Click to expand...
 
cracauer@ said:
If you actively find the Google AI preview in search useful to you, you should consider getting a small subscription to Anthropic Claude. Same thing but Google AI preview is fed by small and fast models, Anthropic's models are full-size. I suppose subscripting to Google Gemini would do similarly better.
Click to expand...
What are the biggest differences between the free vs. paid models you have noticed, especially for development reasons?

Have you experimented with giving the full-size models autonomy to write large amounts code and iterate?
 
I will say for myself, I have a hard time using search engines now, when I'm looking for a solution to a problem, often in the first list are sites with advertising, so we have a situation where sometimes it's better to ask AI, at least it has intelligence
 
I understand that these are neural networks with llm, but earlier, when I googled the error indication of the HP LJ 1200 printer, I found an article even by the manufacturer, and now there are three pages where to buy a printer
 
encryptddthat said:
I would say a chatbot makes information retrieval more specialized yet accessible. The ability to ask follow up queries and probabilistic filtering(the two things that make an LLM) is why chatbots can be much more powerful than search engines, especially for the general population with general type questions.


To top it off, those people look at me crazy when I say I use the free version!
Click to expand...
Yeah, the learning/teaching is excellent and fast. Provides sources as well, if one needs to verify or dig deeper on his own.
My real life skill to scan quickly vast documents for useful info is no longer needed, thank's God.
Also especially handy for code commenting. It is not lazy and thus properly explains lines of code, even trivialities if one is learning the programming language.

The paid models have main advantage of large context window. You can feed it the whole project and it won't choke up so easily. Also it can automate a lot of steps for you, even plan them on it's own. It is usually OK but a little bit off and the correction means having redo all of the project usually....and wasting tokens.
It is similar to human work. Proper Planning Prevents Problems. Measure twice, cut once.
Human guardrails set early are important, especially for non-generic stuff. Which is usually why you do it and not using something already programmed.

I am not sure AI can improve it's game beyond probability machine towards reasoning, logic and broader correlations but so far it is damn useful in the right hands.
A lot of very smart people are not fond of learning proper syntax and catching bugs over sleepless nights, hunched over keyboard. And for me, not so smart and lazy meatbag but with ideas to automate my work, it is a life saver.

And let me tell you, in corporate setting with tons of corporate memo slop, it is huge time saver and already perfect, for hallucinations have no real impact.
 
resora said:
I understand that these are neural networks with llm, but earlier, when I googled the error indication of the HP LJ 1200 printer, I found an article even by the manufacturer, and now there are three pages where to buy a printer
Click to expand...

If you own a HP printer now that is proper advice.
 
You must log in or register to reply here.
Back
Top