FYI, I haven't actually upgraded my backend server yet so if any of you find it being slow let me know. My target response latency is two to five seconds and a maximum of 10 seconds. Anything beyond that and the body language reactions I got from watching people use it degraded.
Please please please tell me if it's longer than you'd expect. I'm still learning a lot about how to scale services properly.
Turns out, since embeddings represent "meaning" instead of keywords, given the right model, this kind of search is somewhat automatically 'internationalized'. There's some discussion about it going on in OpenAI's forums.
It looks like this is an active area of research so I'm not sure what kind of quality you'd get yet but IMHO, it raises some interesting use cases. I came across it for something unrelated but a direct example of how I'd use it.
E.g. Occasionally, I use the 'Translate this page' feature when I end up on a page using a language I don't know, but beyond that specific page (e.g. blog post), I can't do any searches. For the most part, the non-English internet isn't 'accessible' to me. But if I'm understanding correctly, if the search on a website/Google/etc were embedding based, I'd be able to search other-language content even if my query is in English.
Seems like cross-language search and 'Translate this page' combined could be pretty useful to make more of the knowledge on the internet broadly accessible.
For a while now I've wanted a generalized higher dimensional acceleration structure for AI vector embeddings that is conceptually similar to ray tracing bounding volume heirarchies.
What do y'all think. For millions of high dimensional vectors, what new precomputable and incrementally updatable data structures do you think will be useful?
Personally, I'm looking for an efficient nearest neighbour and other efficient 'intersection' tools (e.g. finding a vector most similar to a group of n vector embeddings or finding m pre computed groups that are similar by some distance calculation to a specific embedding).
How do y'all define 'similar music' anyway. I haven't found the perfect playlist creator or music recommender. Spotify's radio has a pretty low hit rate for me.
In some ways, I think there's a market for a multidimensional music similarity search tool or classifier (where playlists are basically classifications).
A few examples. I have a 'rock concert' playlist that maps on style, artist, era/decade, but not tempo (since rock concerts sometimes break up intense songs with less intense ones) and a 'slow lounge' playlist that maps onto instruments, average tempo, tempo variance within song, etc.
What I really want is a way to assign a feel or purpose across a few axes which are not just the typical 'genre'. Something actually objectively measurable like tempo, volume, etc.
Edit: As a side note, I found that country (like Alan Jackson) seems to work better in the car than some other song types. I think it might actually be a frequency thing where country is Darwinism optimized to be audible and enjoyable in old trucks?
Nice thanks lol. I had to do a recent overhaul of this. And I agree, this is not a chatbot. v1 was a chatbot actually, but it wasn't received well for various reasons, one being that you get just one response per question, no peripheral information. I designed it to be more like tweets where you see the thing you looked for, but also get to see related stuff.
This provides a secondary benefit of making hallucinations much much rarer. Try making it lie or make stuff up, it's pretty resilient now. If you find one, let me know; I haven't seen one in a while now thanks to y'all's redteaming.
It's a custom template unfortunately, I can change the colors but it's not modular at all (yeah I know, should have maybe started that way).
For reference, I'm using React + Typescript + Tailwind.
Lol, this is exactly the point. You're more willing to ask what you really want to ask which draws you into a conversation and ultimately makes the person you're asking more memorable to you.
Also, so far, looks like the anti-hallucination stuff is working well but let me know if you run into something unexpected.
Hahahaha that's great. Yeah I see what you did there.
Originally I had mine in chat form but for various reasons, it seemed like an FAQ format with 'sources' resonated better. I could be wrong, but I think that makes OpenAI's custom GPTs an ill fit for the specific problem of networking.
Thanks! They're a part of what I'm using yes but I actually have a pluggable AI system because their API had...we'll say reliability issues. So I can diversify more if needed. I've moved some stuff off OpenAI infra already.
I think there's a lot of work to do to work out the kinks with things like hallucinations, but I think we forget sometimes so much of what we do is statistical in nature already. Your car has a statistical chance of breaking down within x timeframe and while you're on the highway, it's just, low enough that you don't worry about it. I think AI will be a similar thing where we have to get comfortable with how we evaluate and mitigate risks, but like many things, they'll never be 0.
Okay since this is getting way more attention than I thought. Maybe it ends up just being a toy, but if you want one of your own, sign up or email me at [email protected]. It's a no-code solution, and starting takes only a few minutes with a resume/blogpost/autobiography/etc import.
I thought I'd get like, 1 upvote and no comments. Y'all are breaking my question review page. I'll need to paginate it.
Sidenote: I was thinking of making an FAQ of my website as one of these AI profiles too. Maybe I should do that.
The count is up to >1700 questions asked, as a networking tool, I guess it does actually draw attention. Lots of good questions too (and some strange ones), y'all are weird.
This was a fun one to think about. The secret is in where this link is posted. I was thinking through domain name verification and realized that if employers trust LinkedIn, and you can put this link onto your LinkedIn, it inherits that trust.
So basically, if you link it on your resume/LinkedIn, then the consumer of your AI profile should be able to have the same base level of trust in it.
As for the candidate themselves lying, well, you can always do that. In this case, you can actually verify it better than just having to trust a candidate in the moment in an interview. I see these interactive profiles as a way to actually build more trust and credibility between people by giving everyone more to cross-reference.
The candidate gets to be remembered and have the details of their skills and accomplishments shown, the employers get to select with more precision and make valuable interview time even more value dense by having a better/more informed starting point.
Dealing with hallucinations was really finicky. I think I have a way to prevent that now though and ultimately give you control. 'Document search' style things always hallucinated for me so I needed something else. I made another comment on dealing with that but let me know if you have more questions.
The critical mass of users was a weird one to think about yeah. I had two choices, go with breadth first tools (e.g. recruiter tools), or depth first tools (e.g. individual tools to 'market yourself' or just network). I went with the latter first because it would provide value to a single person, just like how my test profile here is providing value to me right now; I didn't want to have to rely on critical mass.
Thanks, yeah! I agree. The import tools was one of the biggest things I had to work on recently.
I needed to balance import everything with no user input (which is prone to hallucinations and doesn't give you an idea of how the AI will answer about you) and asking the user to do too much.
I landed on a halfway point kind of like writing little blurbs about your career that you can add on incrementally. Unlike retraining an AI, you can just add incremental bits.
So the current form is import resume/blog post/etc which generates questions and answer pairs called 'Snippets'. You can add more detail since the resume is usually pretty vague, then officially add these Snippets to your AI.
Once people ask you questions, if there is an answer, it'll use those blurbs. If there is no information yet, it'll say so. You will see all the questions asked of your AI so you can incrementally add Snippets on your Monday night or whatever based on what people had asked your AI profile.
I might change it, but this seems like a good balance between automation, giving you control, and making it incrementally updatable.
Btw, it's all zero-latency, if you make a change, it basically takes immediate effect on your AI. This was an important property to me.
I'm sticking to just URLs for now because they can go anywhere but eventually if it actually becomes a bigger thing, I'd make it easy to integrate into anything (extensions, iframes to put on your personal website, etc).
Also, does anyone else find it ironic that we put LinkedIn links onto our resumes even though it's usually older/more out of date/emptier than the resume. My dream is that a link to your AI profile is on your resume, LinkedIn, etc instead. Give people a way to dig deeper instead of just circularly referencing things.
And yes, doing integrations to your twitter, github, LinkedIn, etc is something I was thinking about too. That being said, you as a candidate might not want all that searchable. I certainly don't. I wanted more control about what was said about me when asked certain questions so I went the import, edit, add route with Snippets (which are like Q/A pair tweets about your career).
Yeah that was REALLY finicky to get right. I have a few ways to prevent hallucinations and I haven't gotten any in the latest iteration even with really crazy questions being asked. (I encourage you to try to break it).
This is partly why I show the source Snippets (Q/A Pairs written directly by the owner) below the summary as a way to verify the information. Kind of like the AI 'showing it's work. It also let's see more about the owner which is a nice side benefit, or maybe the main benefit.
I can also turn off the AI summary part and leave the AI search part. If this becomes a bigger thing, I might give users a way to enable/disable the potentially hallucinogenic part, but it'll be their choice.
Yeah, there are two angles I'm exploring. LinkedIn is a very recruiter focused breadth first search and not very candidate friendly. I think that's why so many profiles are incomplete, non-existent, old, etc.
So I decided to build something that is useful to a single person first, the person that wants to market themselves. This is the first iteration that seems to be working for me in that regard; as a depth first tool.
This is intended as a 'show and tell and get feedback post', but if any of you want to actually try making one, you can reach out to me at [email protected].
I'm intent on making networking just less painful for everyone, in particular, for the individual. If this gets big enough, then I'll think about breadth first tools.
Thanks! It's very early, like, <2 weeks old in its current form, but I'm pretty pleased with it so far.
I can imagine this being a way to pre-warm any cold interactions I might have with new people I meet at work, conferences, etc. More importantly, as something that helps people remember who I am because you know, why network or meet people if they just forget you.
Please please please tell me if it's longer than you'd expect. I'm still learning a lot about how to scale services properly.