Since Bart is now available in Europe I have both options now and problem of choice :) People who have access to both for a while, what AI tool do you mostly use?
You must log in or # to comment.
Removed by mod
This is a bit close minded and reductive when you can see a large number is examples where these tools beat the shit out of search engines, and search engines have been on a precipitous decline in quality for years now. Not talking shit; the privacy concerns are very valid and i dont think there is anything at all wrong with an anti-ai stance in these areas, I just don’t want novices reading this and thinking these tools are “glorified keyboard autocorrect” when some near version of them is undoubtedly the future of both internet search and internet assistant.
Not sure I fully agree. These tool are really ruddy brilliant at certain things (like writing or translating computer code, drafting certain documents) but they are poor at being factually correct. Unless / until they find a way to fact-check themselves I don’t see them replacing search, just complementing it.
Both have option to turn off history saving. In this case they keep dialog for a short time only.
Regarding glorified keyboard autocorrect… I see this a bit different.
Anyways, thanks for sharing your opinion.
Removed by mod
While I am generally inclined to agree with you, I do think AI is getting a bit more advanced than a glorified autocorrect. If you are interested in hearing some examples of cool things it is capable of I suggest listening to a recent episode of This American Life. https://www.thisamericanlife.org/803/greetings-people-of-earth Act One: First Contact is the section I am referring to
“trust us”
no thank you
I still use Google for ~95% of my queries because I like real sources, comprehensive documentation, and not having to read a wall of text when a one-line answer would have sufficed.
ChatGPT is a good replacement for Quora/Stack Exchange for explaining general knowledge stuff like other languages’ grammar and simple science, as well as finding authors/books/movies from descriptions when you’ve forgotten their names.
Bard is… kinda dumb. I gave it a few chances, but it was nothing compared to ChatGPT’s free tier.
Definitely agree that ChatGPT is often waaay too wordy.
Tried both but honestly haven’t found much of a proper use for either.
I still don’t think it’s much more of a novelty. From what I’ve used it really feels like you can see the training data in all of the answers, which obviously, but like if I ask it to write a cover letter it feels like it’s some cover letter it trained on more than mine
It’s much better than novelty if you learn how to use it. I routinely use it to write scripts for Google apps scripts, bash, etc. I’m using it to help with rust libraries that don’t have much documentation. I’ve used it to research camera gear that I don’t know much about. And once you get some information from it you can then go Google. Google has gotten so bad lately that it’s not hard for ChatGPT to beat it though. If you’re using 3.5 I highly recommend stopping that nonsense though. There’s zero reason to use 3.5 and every reason to use 4.
Neither because AI is bullshit
Just tried out Bard with what I thought was a not to exotic question.
It made up a feature for a library, then lied about when it was added, and finally conocted some imaginary source code for it.
It’s seriously a mystery for me how anyone gets a net benefit from using these LLMs when you have to second-guess literally everything they output.
I use chatgpt for generating ideas for my DND campaign, and for coming up with examples of an idea that I can use as a jumping off point for software development.
Other than that the only LLM I use is GitHub copilot.
Same here, although since OpenAI disabled bing search I occasionally need Bing since it knows better about the new Godot 4.
I find Phind.com quite useful as an IT person.
I have never heard about it. What are the benefits over generic chatGPT?
None, of course.
I’ve been using FastGPT from Kagi search, impressed that it is starting to give sources for what it says.
ChatGPT is my programming rubber ducky for general stuff
I have not had a chance to try Bard. Since ChatGPT came first, I’m still experimenting with its capabilities to see how it can assist me in my workflows. Since we’re on this topic of LLMs, I do suggest anyone using them to read up on “prompt engineering”. One drawback of these models is the generation of ‘hallucinations’ or false information. They work by imitating the next ‘token’ (think autocomplete but with attention on context)–but there’s no fact checking, no reflection, no internal monologue. Better prompting helps avoid these cognitive pitfalls of LLMs and will help produce better/more accurate answers. Prompting LLMs is currently an active area of AI research so a lot is still unknown. However, you can definitely find some pointers and best methods on prompting that help you become a more effective user of these tools.
ChatGPT writes my cover letter first drafts for me. It’s pretty good. I fix the draft, obviously, add things, restructure, but it’s way better at boilerplate corporate newspeak than me. Haven’t tried Bard yet, though.
Neither, but I plan to try PrivateGPT soon. I’m curious about feeding it some CS books and testing the outputs to questions. I may also try feeding it transcriptions of CS lectures. I’m curious if it can be useful when I get stuck on stuff while learning on my own
Neither. I have played with huggingchat.
Haven’t tried bard but use ChatGPT to write/debug scripts and SAP stuff. Also asking it when I have simple but technical questions.
I am also downloading and running the latest models in the local LLM space every 2-3 weeks, just waiting for the point at which they finally take over gpt3.5 at which point I’ll probably not touch ChatGPT again.
Which ones are your downloading and running?
https://github.com/serge-chat/serge
The easiest to set up that I have found and it supports a large library of models. I can not say that I have had amazing results but it is a lot of fun to play with.
Today I tried TheBloke/WizardLM-13B-V1-1-SuperHOT-8K-GGML in koboldcpp for a short bit but didn’t actually test the 8k context. I just love seeing them evolve and get a bit better every time I download a new one.
Interesting, thanks.
