I've been working on my own personal-assistant for a couple of months, just connected it to telegram so I can reach it from anywhere. Can "talk" to my calendar, run commands on my home computer, etc. It also has its own memory, so doesn't need huge prompt windows (I'm running a couple of fine-tuned curie models btw). Now I've been giving it API access to a bunch of stuff to increase its capabilities.
What do you use for the memory ? Can you explain a bit more ?
mysql database, with a column to store the embeddings vector (yeah i'm not so fancy to be using pinecone :P)
Yeah, I gave it a simple task of encoding a secret message in a sentence by using the first letter of every word. Hello = "Hey everyone, lick less onions". I worked with the prompts for over an hour to try to get it to complete the task, and while I did have some success, it really struggled to reason about the task or provide a valid response. If it can't even reason about a child's game, I can imagine it struggles…
I think the most interesting response I've gotten was one where gpt-4 noticed halfway through a response that it had made an error, apologized and then put the corrected sentence. When I queried it, it claimed it generates the response a token at a time, and could not back up when it realized the message was incorrect, but I don't know enough about how the tech works to ascertain the veracity of the statement.
That's exactly right. It can see what it's returned but can't edit it.
> I don't think it's actually reasoning about things at all, just providing a statistically plausible response to prompts. It turns out that humanity’s problem might not be that AIs can think but rather that humans believe that AIs can think. One might even go so far as to say there’s a real danger that we hallucinate that AIs can think, to our detriment.
We don’t actually know what “thinking” is, though, so I’m not sure it’s possible to say “this model can’t think”.
> We don’t actually know what “thinking” is
How about: thinking is to make sense of the world (in general) and decide how to respond to it.
Since it's a transformer model, I use it to... "transform" the data.
- Change raw citation into bibtex entry
- Fix spelling mistakes
- convert csv file into json (though you can do it without employing trillion parameter model)
etc.
I find ChatGPT most helpful as a "what's that called" tool. A lot of my queries are finding/confirming the right idiom when writing something, or getting a specific name out of a vague description (JavaScript concepts, shell commands, CSS selectors). Search engines with SEO are so reliant on keywords that it often feels like I'm suggesting answers rather than asking questions - it's so refreshing to be able to just a…
It's replaced Google for the "what's that called" tool. Google used to be good at that task, but it's sucked for the last four years or so. Whenever they gave up on search and leaned into sludge content plus ads. ChatGPT is better than Google ever was anyway.
GPT is better than Google for something like, 'how do I implement nested blog post comments in Flask'
And much worse than Google for, "site:miguelgrinberg.com nested comments"
Unfortunately I don't know the equivalent of Miguel for everything I need to know, so on average I suppose GPT is better. However it also means that you may never discover the Miguel of your domain!
i think most people will just keep programming the way they do, and the AI hype will mostly die down. People have been saying that C++ is dead for decades, yet here I am writing code in it with a big community of others who do, too.
I think this is the problem. When people talking about c++ is “dead” ,at that time they meant 70% people using to perhaps 5% . Just like we says after industrialization,making clothes by hand is dead . It is irrelevant that there are still some people making clothes by hand . When AI the main way to code and remove 90% of coding jobs. It is also irrelevant to state that there are still 10% people still coding
It doesn't have to be job or career related to be relevant.
So on my coding problems I haven't had much luck. It doesn't seem to know Bazel, the Rust code I asked about was completely hallucinated, but it did solve a problem with Azure DevOps I had. I think if the training set did not contain enough of something it can't really think of a solution. What is really nice though it's as you say the refinement of questions. Sometimes it's hard to think of the right query, maybe yo…
This could mean the future goes one of the two ways. Engineers get lazy and converge on using only programming languages which AIs understand or have been trained on, or we forget about this waste of time and work on more important problems to solve in society other than the lack of an AI to be our crutch. Sadly, I think the former is more likely.
Programmers are per se lazy, at least that is, what i always thought, that it is mostly about automation. With spending little time on survival, we get the time to work on more important problems. Whatever those are. It is not an either or, that is what i try to say! :)
I would like to use GPT for engineering problems at work, but it’s just not practical for me since I work on a large internal codebase. It is very rare for me to need self-contained code solutions on the order of <100 lines.
For being productive? Not at all for the most part. I haven't really found anything that I do that I can punt to ChatGPT. I guess I could have used it to help me write this response, but what would have been the point? The little bit of time I spend messing with it (and Bard now that I have access) is mostly just for fun; trying different jailbreaks and creating ridiculous scenarios and seeing what kind of reaction I…
I'm pretty much the same. I don't find significant productivity gains from using it - maybe because I have a specific way of doing things already. For instance, I know it's better for me to understand React/whatever framework rather than letting chatgpt write the react state/reducers/etc and all that stuff. I can definitely use it for emails and have used to simplify exec emails in my company but that's just it.
Understanding a topic and letting chatgpt do repetitive and simple boilerplate stuff are not exclusive.
What do you use for the memory ? Can you explain a bit more ?
mysql database, with a column to store the embeddings vector (yeah i'm not so fancy to be using pinecone :P)
To keep the "memory", do you pass the embeddings along with the new text prompt in an API call? How do you combine embeddings and text prompts? I don't know much about this, sorry if the question sounds silly.