Ask HN: Is anybody getting value from AI Agents? How so?
I saw a lot of initial buzz about the promise of agent based workflows and it seemed to be the obvious way to get LLMs to the edge of decision making and leverage many specialized models. It seems the chatter has died down but there are growing projects out there in the space. Before I invest the time to explore and work with tools I’d love some feedback from the community and if and how they are being used.
158 comments
[ 2.7 ms ] story [ 177 ms ] threadThat said, I’m very bullish on agents overall though and expect that once they get their assistants behaving a bit more predictably we will see some cool stuff.
It’s really quite magical to see one of these think through how to solve a problem, use custom tools that you implement to help solve it, and come to a solution.
or was GP implying lot of people are gettin funded to build agents in a thread where the consensus is they dont work well enough for people to pay for
Legal angles here also also super interesting. There’s a growing body of scenarios where companies are held accountable for the goofs of their AI “assistants.” Thus we’re likely heading for some comical train wrecks as companies that don’t properly vet this stuff set themselves up for some expensive disasters (eg think the AI assistant doing things that will get the company into trouble).
I’m bullish on the tech, but bearish on the ability of folks to deploy it at scale without making a big expensive mess.
Right now the USCO says if you use a large model that used copyrighted material to train it, you cannot copyright the generated art. This applies to not just art but might be for code as well. So I wonder what the legal liabilities are say for a publicly traded company to use Midjourney to generate content that is also copyrighted.
ex) it is possible to generate movie characters that we instantly recognize if you use non-english words which pretty much nail in the coffin for safe harbour status granted under DMCA
Unless the Copyright Office has come out with a newer rule, it says if you use a model that does the usual creative parts of making the output you can’t copyright the output, because it is not a work of human authorship. (And that if a work is mixed AI/human, the AI use must be disclosed in copyright registration to avoid copyright protection being applied to elements that are not human work.) “Large model” and “copyrighted material used to train it” are not the issues.
The past is so often proven wrong and was so much less civil, we can never actually ask James Madison if we did right by his notion the future owes the past. We can only decide for ourselves we align with the long dead’s philosophy.
Where is the value in so carefully conserving in such detail the past? I’m not saying we end history as a field of study but as an obligation of daily life.
The most important lesson of history, imo, is prevent nation states or whatever might take their place from having the ability to wage wars. Most of their philosophy is indecipherable given our life experience is vastly different than theirs. Would they see this world and come to the same conclusion they did?
https://www.theguardian.com/world/2024/feb/16/air-canada-cha...
The first dominoes to fall will be art, music, and film. These don't require perfection and can be iterated on by the creator. The tools become a method in the process.
Everyone in the agent space is bumping around until the next big innovation that actually unlocks the technique sparks a race to productize. Maybe some will get lucky and have a fast pivot. Or maybe they'll run out of funds before the big breakthrough.
Sometimes I get they are already sitting on some ground breaking stuff and slowly releasing it to test our responses and get feedbacks.
I'm certain they will be reading this thread but if one theme repeats itself is the lack of trust in the output of the agents and the companies creating it.
Absolutely. These agent startups don't have PMF and haven't solved anything yet. They're playing in the kiddie pool while Microsoft is placing chess pieces with a GDP-level moat (which is frankly terrifying).
The idea does not suck. On paper it's great. Just tell an "agent" what you want and off it goes to get it. And it's possible. LLMs open the interface to search planning and selection algorithms that have been around for 40 years and are mature. You could have this tomorrow.
The assumption is that people want it.
Tech business has come a long way exploiting people's vices, specifically laziness of thought we call "convenience". But at heart, tech is still seen as a tool, to empower people, to give them agency.
Agents subtract agency.
> It's too early.... the big breakthrough.
Hoping for progress against human psychology seems a fool's errand.
With postmodernism art deconstructed its duty and without duty society does not grant rights. It is a big problem because art is often visionary about the future. We don’t know that we even have a future without vision.
There isn't enough art in the world. I've spent the last two decades making films the photons-on-expensive-glass method, and it's a pain in the ass.
I can start new software projects of scale easily. I can't do that with art. It's capital intensive, logistics intensive, and requires too many people in low-autonomy roles.
It really sucks the fun out of storytelling when you're chasing down location rights, showing up at the prop house at 7 AM, arranging for catering, etc.
> We don’t know that we even have a future without vision.
As a creator, I've dreamed about a better way before GenAI was even on the radar. I'm not the only one. Existing processes suck and too much of the work isn't creative at all.
As a consumer, my needs are barely being met at all. I want a show about steampunk vampires in space. I want a biopic film exploring Reimann, but told as a musical. I have creative notes for Benioff and Weiss, and I'd rather put those together for myself and my friends than echo words into the void.
I want so much more than the canned limited selection I have available on Netflix and Criterion. It barely whets the palate, and my appetite is completely unsated. The closest I've ever gotten is performing in improv theater and exploring the worlds I want with the people I feel comfortable creating around. But that's only part of the experience I want. I want so much more. The art we have today is a pale shadow of the mind's unlimited canvas. A projection onto a caveman's wall.
You can protect a vision of a priestly class of artists from the printing press all you want. I'm tired of living in the dark ages when we're sitting at the precipice of so much more. Their jobs won't go away any more than wedding and event photographers suffer in the advent of digital film.
If anything, the artists will be the best primed to take advantage of the new alpha. Free of studio meddling, they can build their own audiences that they own without the chains of brand guidelines. Vivziepop, but a million fold.
I saw a figure recently that said 80% of consumer film, games, and media originates in the US. Think about the rest of the world. So much culture and perspective that we should all share in remains unseen, and we're left with the lens of US media giants.
Think, too, of all the dreamers lost in opportunity cost. There are a billion stories that die silently in the minds of dreamers because we couldn't help them. And the world is all the worse for it.
If a collective shoulders that duty then we do get a printing press priesthood. If individualists shoulder the duty to the collective then we should get something different. Perhaps productions like Helluva Boss do have some kind of post-deconstructionist duty of their own design. If that is the case then society should reciprocate with granting rights different from what the priesthood would accrue.
Let's say if you try to make a Family Guy episode using generated video, the characters would come out slightly different in every scene. You could generate 100 outputs for every scene and try to pick the best one but it still won't be very good.
Another one is listening to many social media (e.g. Twitter) posts to sense if there is a business opportunity. SDRs scan the results in an Slack channel manually but based on these signals.
Finally, this is now a workflow but we did this [1] that is a piece in our work.
[1] https://news.ycombinator.com/item?id=39280358
It’s actually quite awful. It’s obvious the text is LLM generated because of the verbose, generic writing style. It communicates clearly but without substance. Not gonna lie, I secretly judge these people.
If your math worked out, then humans couldn't work either.
If you already have the business process robustly defined as code, then the utility of LLM is unclear. The value prop of LLM is in fuzzy business processes like parsing arbitrary helpdesk tickets.
LLMs didn't invent stochastic process steps.
I also often correct my own mistakes based on clashes with reality - I don't just become more intelligent the second time.
The paper is currently only available at Arxiv (ie not yet peer-reviewed), but given that it is Papadimitriou, I would be inclined to believe the results.
https://arxiv.org/abs/2402.08164
https://www.klarna.com/international/press/klarna-ai-assista...
I haven't used Klarna, but a few other products I've used are unsupported now because human support were replaced with AI "support" that is completely useless.
- GPU bill is $200/month for 21,000 designs per month or about 1¢ per render (no character training like Photo AI helps costs) - Hosted on a shared VPS with my other sites @ $500/mo, but % wise Interior AI is ~$50 of that
+= $250/month in costs
It makes about $45,000 in MRR and so $44,730 is pure profits! It is 100% ran by AI robots, no people
I lead the robots and do product dev but only when necessary"
https://twitter.com/levelsio/status/1773443837320380759
And neither do most restaurants either; what's to prevent someone from getting a service and then a full refund?
I’ve found a lot of these numbers from people selling passive income methods are extrapolated.
All of that just to sell a book?
I've seen a lot of success come from AI sales agents, just doing basic SDR style work
We're having some success automating manual workflows for companies at Skyvern, but we've only begun to scratch the surface.
I suspect that this will play out a lot like the iPhone era -- first few years will be a lot of discovery and iteration, then things will kick into superdrive and you'll see major shifts in user behavior
You mean any.
You can fire a human knowledge worker for not doing their job correctly, but what are you gonna do when you only have LLMs and realize they can't do their job correctly?
Basically the same what we have now, except the grunt workers will be replaced by the machine.
Not junior, unthinking. It's like outsourcing/offshoring. You're getting the equivalent of someone who doesn't care about what they're doing, not somebody inexperienced. I'm not saying it doesn't have a place, just commenting on the framing.
- The pythonrepl or llm-math agent not being used when it should be and the agent returning a wrong or approximate answer.
- The wikipedia and webbrowsed agents doing spurious research in an attempt to answer a question I did not ask (hallucinating a question, essentially).
- Agents getting stuck in a loop of asking the same question over and over until they time out.
- The model not believing an answer it gets from an agent (eg using a Python function to get today's date and not believing the answer because "The date is in the future").
When you layer all this on top of the usual challenges of writing prompts (plus, with Python function, writing the docstring so the agent knows when to call it), wrong answers, hallucination, etc, etc, I'm unconvinced. But maybe I'm doing it wrong!
It's true that have to wrestle a lot with them to get them to do what I want for more complex tasks... so they are great for certain tasks and terrible for others, but when I'm in Xcode, I dearly miss vscode because of Copilot autocomplete, which I guess is an indication that it adds some value
One unexpected synergy has been how good GPT4 is at explaining why my rust code is so bad, thanks to the very verbose compiler messages and availability of high quality training data (i.e. the great rust code in the wild)—despite GPT4 not always being great at writing new rust code from a blank file.
Part of me thinks in the future this loop is going to be a bit more automated, with an LLM in the mix... similar to how LSPs are "obvious" and ubiquitous these days
On an unrelated note, I also wrote a small python script for translating my Xcode project's localizable strings into ~10 different languages with some carefully constructed instructions and error checking (basically some simple JSON validation before OpenAI offered JSON as a response type). I only speak ~2 of the target languages, and only 1 natively, but from a quick review the translations seemed mostly fine. Definitely a solid starting point
Specifically on using AI for coding, I wrote about different levels of AI coding from L1 to L5, we are still at L2/L3 stage for mature and production ready tech. Agents are L4/L5:
https://prompt.16x.engineer/blog/ai-coding-l1-l5
That line of reasoning has held true across basically every project we’ve touched that tried to incorporate LLMs into a core workflow.
Then they should be great for making fast food, staffing amusement parks, and seasonal farm labor.
They don't seem to be good for those things either.
[Edit to add: the value high schoolers bring to jobs through non-cognitive abilities, which AIs lack.]
I feel like these architectures built on top of last gen LLMs are mostly useless now.
The current gen jump was significant enough that creating a complex chain of thought with RAG on last gen usually is surpassed by 0 shots on current gen.
So instead of spending time and money building it it's better to focus on 0-shot and update your models to the latest version.
Feeding LLM outputs into other LLM inputs IMHO just will increase the bias. Initially I expected to mix and match different models to avoid it but that didn't work as much as I expected.
It depends a lot on your application honestly.
All that's left is for someone to bundle it all up into a nice package, and we'll be in the future.
* I was able to make a simple AI agent that could control my Spotify account, and make playlists based on its world knowledge (rather than Spotify recommendation algos), which was really cool. I used it pretty frequently to guide Spotify into my music tastes, and would say I got value out of it.
* GPT-4 worked quite well actually, GPT-3.5 worked maybe 80% of the time. Mixtral did not work at all, aside from needing hacks/workarounds to get function-calling working in the first place.
* It was very slow and VERY expensive. Needing CoT was a limitation. Could easily rack up $30/day just testing it.
My overall takeaway: it's too early: too expensive, too slow, too unreliable. Unless you somehow have a breakthrough with a custom model.
From the marketing side, people just don't "get it." I've since niched down, and it's very, very promising from a business perspective.
[1] https://konos.ai
or are we at the mercy of hosted models?
I do think AI "agents" (or blocks as I like to think of them) unlock the potential for solving unstructured but well-scoped tasks. But it is a block of unstructured work that is very unique to a problem, and you are very likely to not find another problem where that block fits. So, trying to productize these AI blocks as re-usable agents is not that great of a value prop. And building a node based workflow tool is even less of a value prop.
However, if you can flip it inside out and build an AI agent that takes a question and outputs a node based workflow. But the blocks in the workflow are structured pre-defined blocks with deterministic inputs and outputs, or a custom AI block that you yourself built, then that is something I can find value in. This is almost like the function calling capabilities of GPT.
Building these block reminds me of the early days of cloud computing. Back then the patterns for high availability were not well-established and people that were sold on the scalability aspects of cloud computing and got onboard without accounting for service failure/availability scenarios and the ephemeral nature of EC2 instances were left burned, complaining about the unfeasibility of cloud computing.
That rings useful to me. I find it hard to trust an AI black box to output a good result, especially chained in a sequence of blocks. They may accumulate error.
But AIs are great recommender systems. If it can output a sequence of blocks that are fully deterministic, I can run the sequence once, see it outputs a good result and trust it to output a good result in the future given I more or less understand what each individual box does. There may still be edge cases, and maybe the AI can also suggest when the workflow breaks, but at least I know it outputs the same result given the same input.
- Navigation: Detect navigation elements and handle actions like pagination or infinite scroll automatically.
- Network Analysis: Identify desired data within network calls.
- Data transformation: Clean and map the data into the desired format. Finetuned small and performant LLMs are great at this task with a high reliability.
The main challenge:
We quickly realized that doing this for a few data sources with low complexity is one thing, doing it for thousands of websites in a reliable, scalable, and cost-efficient way is a whole different beast.
The integration of tightly constrained agents with traditional engineering methods effectively solved this issue for us.
First, we know these AIs are trained with data from the general Internet, and that data is vast.
Second, the general Internet contains owner manuals and support forums for practically every active product there is, globally. These are every possible product too: physical products, virtual products like software or music, and experience products like travel or education. Between the owner’s manuals and the support forums for these products there is extremely deep knowledge about the purpose, use and troubleshooting of these products.
Third, one cannot just ask an LLM direct deep questions about some random product and expect a deep knowledge answer. One has to first create the context within the LLM that activates the area(s) of deep knowledge you want your answers to arise. This requires the use of long form prompts that create the expert you want, and once that expert is active in the LLM’s context, then you ask it questions and receive the deep knowledge answers desired.
Fourth, one can create an LLM agent that helps a person create the LLM agent they want, the LLM agent can help generate new agents, and dependency chains between different agents are not difficult at all, including information exchange between groups of agents collaborating on shared replies to requests.
And last, all that deep information about using pretty much every software there is can be tapped with careful prompting to create the context of an expert user of that software, and experts such as these can become plugins and drivers for that software. It's at our finger tips...!