hckrnws

First off, congrats on the funding and the progress so far!

I’ve seen a a couple of start ups pitching similar ideas lately - platforms that use AI personas or agents to simulate focus groups, either for testing products or collecting user insights. I can see the appeal in scaling audience feedback, reducing costs, reaching demographics that are traditionally hard to access.

That said, this is one of the areas of AI that gives me the most concern. I work at a company building AI tools for creative professionals, so I'm regularly exposed to the ethical and sustainability concerns in this space. But with AI personas specifically, there is something a little more troubling to me.

One recent pitch really stuck with me, in this case, the startup was proposing to use AI personas for focus groups on products and casually mentioned local government consultation. That's where I think this starts to veer into troubling territory. The idea of a local council using synthetic personas instead of talking directly to residents about policy decisions is alarming. It may be faster, cheaper, or even easier to implement but it fundamentally misunderstands what real feedback looks like.

LLMs don't live in communities. They don't vote, experience public services, or have lived context. No matter how well calibrated or "representative" the personas are claimed to be, they are ultimately a reflection of training data and assumptions - not the messy, multimodal, contradictory, emotional reality of human beings. And yet, decisions based on these synthetic signals could end up shaping products, experiences, or even policies that affect real people.

We're entering an era where human behaviour is being abstracted and compressed into models, and then treated as if it's a reliable proxy for actual human insight. That's a level of abstraction I'm deeply uncomfortable with and it's not a signal I think I would ever trust, regardless of how well it's marketed.

Would be curious to know what your approach is to convince others that may also be skeptical or not want to see this kind of tech being abused for the reasons listed above?

Thank you! We 100% agree. My research back in Cambridge was on misinformation, so we take the danger of misuse very seriously even as a tiny team of 3 people right now. As a social science researcher, one big challenge we faced was just how difficult it was to run experiments - it's quite unethical (and impossible) to have 100k people under policy A and 100k under policy B, so as a result, we as a society struggle to find the "golden path" with big issues like misinformation, climate change, or even everyday economics.

That's what motivated me to start researching in the area of creating "Artificial Societies" - first as an academic project, now as a product everyone can use, because I believe the best way to build a new technology is to try to make it useful for as many people as possible, rather than reserving it for governments and enterprises only. That's why unlike other builders in this space, we've made it a rule to never touch defence use cases; that's why we've gone against much business wisdom to produce a consumer product that anyone can use, rather than going after bigger budgets.

We totally agree that synthetic audiences should never replace listening to real people - we ourselves actually insist on doing manual user interviews so that we can feel our users pain ourselves. We hope what we build doesn't replace traditional methods, but expands what market research can do - that's why we try to simulate how people behave in communities and influence one another, so that we capture the ripple effects that a traditional survey ignores because it treats humans like isolated line items, rather than the communities we really are.

Hopefully, one day, just like a new plane is first tested in a wind tunnel before risking the life of a test pilot, a new policy will also first be tested in an artificial society, before risking unintended consequences in real participants. We are still in the early days though, so for now, we are just working hard to make a product people would love to use :)

But "artificial societies" are only possible with AGI, not with LLMs. These are not reasoning engines. They do not think or have values or care or worry.

Someone must have a wild-ass theorem about whether or not consciousness is representable as some distribution over possible realities. But yeah, I agree this feels like taking a huge step towards fewer and fewer people having agency in their own (real) lives.

I'm certain Big [insert industry] will gobble this kind of thing up.

Great idea and execution.

Is there any way to know if you ran a simulation of your message on Hacker News, and whether the actual outcome matched what your simulation predicted?

I'm a stats nerd, so I'm just here for the data.

Exactly my concerns as well. If we're indeed heading toward “ask AI first, humans later” model there's potential for a slippery slope—one that could be exploited depending on which regime happens to be in power. If politicians or special-interest groups can manipulate or curate AI-generated “opinions,” they could present those biased outputs as if they were genuine reflections of their constituents’ views. Over time, the line between authentic public sentiment and engineered AI propaganda could blur, undermining informed democratic debate.

See https://en.wikipedia.org/wiki/Franchise_(short_story)

I’m curious if you think this is somehow worse than a bunch of 50-year-old mostly white dudes sitting around making product decisions based on their impressions of college students these days?

You make a strong case for why this approach is imperfect, but I’m not seeing how it is worse than current practice.

I read the accuracy report and I'm yet to find on what basis is your accuracy score being built on. Is it the number of personas that re-post, like, comment, see, all of the above?

I think you guys might be onto something but I'm still skeptic as to whether you are the most accurate (on whatever metric). It's not surprising that you beat a survey of experts, or straight out of the box commercial LLMs.

I'm more interested in seeing how your model performs against purpose specific models that are currently industry standard. Unless you're making the claim that you're the first service to predict content engagement?

Nice work, clean demo. Who is your target buyer?

This seems like it could be useful for product discovery (“what are these people complaining about?”), content marketing (“how will my twitter followers react to this blog post?”), and other… reactionary… activities. But what about GTM and lead-gen? Can you ask it “who has job title of CISO, within two degrees of connection to me, working at a company with at least 500 employees that is SOC2 certified?”

I think you need to focus on a target buyer and make sure you nail their use case, or you risk wasting time on a really cool product that kinda/sorta does everything.

What’s your differentiator? Are you in the business of data gathering and curation? Or do you enhance some existing targeting data with “talk to my audience?” These are two distinct product development paths… either you invest in sophisticated data scraping, or you focus on “bring your own audience.” Most companies already have this sort of data on their customers and prospects – how can you meet them where they are?

The other problem is garbage in, garbage out. This product is only as useful as the data you can gather (or that your customer brings to you). A list of emails and names isn’t much on its own. You need the data generated by those people. Maybe you need to partner with data brokers to enrich audience data with social media profiles. Or maybe you leave it up to your customer – let them upload all their support tickets (ZenDesk) and sales calls (Gong) to your software so that they can “talk to their customers.” (Hint: maybe you should partner with Gong, and similar companies who already have this data, to provide this feature to their customers. White-labeling this product might be your fastest path to market.)

But more existentially… is “opinions” the most important aspect of your customer that you want to simulate? And if so… why do you need a _network_ of customers for this? It seems like two disconnected ideas. An “audience” might be a group of people that you only know to be associated because of their shared subscription to your product. Or they all follow someone on Twitter. Or they’ve all written an HN comment with “trivial” in its text. Does the _network_ aspect actually matter, at all?

And if the network aspect is important… why? Is it because you want to discover a new audience? In that case, are you focusing on the right value by simulating the current audience? Or should you be focused more on features for expanding/enriching/discovering “similar” profiles?

I think you’ve got the basis of something really cool here, but you need to figure out your identity and core competency, or you risk doing a bunch of marginally useful stuff kinda well.

Did you see the recent HN launch of Sumble? I see some overlap and similarities between your products, and I’d suggest reaching out to them in case you can work together…

edit: Just saw you went to Cambridge… I live here, if you’re ever around and want to grab coffee. I’m “Miles Richardson” on LinkedIn (and uh… in life). Feel free to message me… I spent five years on a startup that never hit product/market fit, so I’m always happy to point out the hazards…

Chatbots are the wrong approach. You should construct personalities and use the various traits and components of the marketing message as considerations for a utility AI that determines what action it will take when presented with the content. Perhaps the only place you'd use LLMs is the score the marketing message on various attributes, but that could also just be done through cheap sentiment analysis and classifiers.

This is very cool. I tested it with a specific linkedin post and it was pretty much spot-on (the reaction estimate, that is). Can't wait to try the optimized versions in the future.

Do you ensure you have enough personas in the desired target area to get enough survey responses? Are there demographics you are not able to simulate at this time?

Seems like it would be impossible to simulate anyone who isn't chronically publicly online?

I'm chronically online but I have very few public profiles anyone could glean anything from. (And even the one I'm posting this with here is in the queue for deletion... supposedly... I should probably check on my request.)

It’s very rare that dang actually satisfies deletion requests as it leaves gaping holes in the historic HN threads. You see this constantly on Reddit and it’s a major problem. For that reason, it’s best to follow the policy of not posting something online that you don’t intend to be permanent.

My intent wasn't to redact my comment history (I also value that permanence), just remove attribution/profile and any ability to log in so I stop being tempted to comment further.

just post your password, I'm sure we can get you locked out

I just tried it out. Very impressive! I like the user interface and user experience.

I'm a Director of Market Research, so I did a test:

- One email headline test for a service I'm considering developing.

  -- got a score of 88 - "Exceptional".

  -- got a score of 81 - "Very high"

Feel free to DM me - I really like what you're doing and would be happy to discuss.

Further thoughts: I was asking to simulate some very specific target markets. There's no reason society.io should be able to simulate every possible market. Possibly when someone describes their target market, you can respond with an assessment of whether that is within the set of markets you currently support.

For what its worth I think this is super cool.

As a serial wantrepreneur I know what the letters SEO and A/B testing are but little else about this side of business. As a total novice, the ability to run a bunch of experiments quickly and cheaply, even if there isn't a 1:1 mapping to the real world, seems like a huge win.

To be honest I think I feel this way about most things ai/llm right now. I am ok taking the train to the stadium and walking half a mile over taking an uber to the front door and waiting in traffic.

I use AI to create customer avatars representing potential buyers of a product I may create (based on existing competitors and their customer reviews). I then use those customer avatars to help design the product.

I love the idea of going from "AI generated customer avatar" to "simulated real people". It would help add depth to the customer avatars, and lead to better product design.

I tried creating a society around products that I sell, but it looks like the "real-world data" is pulled from LinkedIn? I'm not necessarily targeting business people.

Thank you for trying us out! Yes, most of our personas are built from social media profiles + other deep research of publicly available data. For this reason, our customers have made the most of us by simulating professionals who are otherwise really hard to survey.

Define a new series in p: p.mySeries = new BrownianGenerator(start, mu, sigma) p.mySeries = new BrownianGenerator() // With default params

Combine and transform series: p.ma = p.mySeries.sma(20) // Simple Moving Average p.ema = p.mySeries.ema(10) // Exponential Moving Average p.emaCustom = p.mySeries.ema(10, 0.3) // EMA (specify alpha) p.diff = p.mySeries.subtract(p.ma) // Difference p.sum = p.mySeries.add(p.ma) // Sum p.prod = p.mySeries.mul(p.ma) // Multiply p.prod = p.mySeries.mul(8) // All arithmetic operations work with numbers p.absdiff = p.diff.abs() // Absolute

Chain series:

p.diff10 = p.mySeries.subtract(p.ma).mul(10)

Render series in the chart: show(p.mySeries, p.ma) // Show selected series showAll() // Show all series in p

Congrats on the launch! I've been watching for products in this space and this looks really nice. The UX is really well thought through. Great product demo.

Hadn't seen that paper, thanks for sharing it. This is the one I see cited most often that's got some similar vibes: https://arxiv.org/abs/2411.10109

Thank you so much!! Indeed, we were very inspired by the Stanford team's work as well :)

This is a fascinating project! It's exciting to see the application of multi-agent simulations to a real-world problem like optimizing marketing messages. Your approach of grounding the AI personas in public data and focusing on social influence is very insightful.

I recently authored a paper on a related topic that might be of interest, particularly concerning your challenges with accuracy and long-term simulation stability. Our work, "SALM: A Multi-Agent Framework for Language Model-Driven Social Network Simulation," introduces a novel framework for integrating language models into social network simulations.

I specifically focused on achieving long-term temporal stability in multi-agent scenarios.[2] A couple of our key contributions could be relevant to your work:

I developed a hierarchical prompting architecture that enables stable simulations beyond 4,000 timesteps while significantly reducing token usage.[3]

To address memory growth and maintain agent consistency, I implemented an attention-based memory system that achieves high cache hit rates with sub-linear memory growth.

I also established formal bounds on personality stability, which could be a useful concept for ensuring the accuracy and predictability of your AI personas over time.

Our validation against SNAP ego networks demonstrated the capability of our framework to model long-term social phenomena with empirically validated behavioral fidelity.[3]

It seems like there's a lot of synergy between our research and your product. I'd be very interested to hear your thoughts on our paper and explore if any of our findings could be beneficial for the continued development of Artificial Societies.

Here's the link to my paper: https://arxiv.org/abs/2505.09081

Doesn't this defeat the purpose of actual field research?

If it isn't based on ACTUAL buyers who have ACTUAL input, what is it really doing? Great job at creating something, but at the same time, it feels kind of unnecessary.

You're essentially telling people what you THINK they'll want.

Not necessarily.

You can make it collect personas from actual users who actually interact with you on social media, for example Facebook page fans or x.com followers.

If the models are built from those actual users and their public social graphs, that gives a lot of data points to the inference engine about their demographics as well as their interests.

No idea if the results are statistically relevant, but it might bring good enough results at a fraction of the time*cost of an actual study (that probably won't be statistically accurate either anyway).

That will generate comments which are statistically representative of how people interact with you online, not actual people who actually might buy your product's opinion that this design looks worse than the old one or that this ad is misleading or offensive

If any of the ai caricatures had android phones running Firefox they'd tell you that page is really rough to scroll, particularly the first half

so sorry about this! fixing it now. we simulated how our content would land, but alas couldn't test the site before it was built :`)

[flagged]

Please don't be a jerk on HN, especially in response to someone's work. Nobody gets everything perfect the first time.

https://news.ycombinator.com/newsguidelines.html

I ran some tests on a similar concept using LangGraph. Unfortunately, I think while the results are meaningfully different from a foundation model, they don't supplement actual real-world data yet and don't provide sufficient diversity of thought and opinion between cases. For instance, asking 1,000 people you get a different response every time because each person is slightly unique, but with the LLM it is probabilistically different, not because of the slightly unique differences and for Meta reasons.

Is there a comparison between a campaign simulated here and it's real world performance?

Congrats, this is amazing and clearly I will want to use this for Linkedin, how can we get in touch to get more details in order get an active support for the first situations, for example, right now we’ve been using agencies in buffer to accommodate our Linkedin strategy

The problem with this (and for that matter every other scientific and technical advance) is that if it works well - regardless of whatever high minded upsides exist initially - it will also have corresponding downsides.

In this case, it's a technology that is hoping to significantly enhance the ability of people with the money to sell whatever they like to targets they have modelled.

Leaving aside the stalky model bit, the thing that should give pause is where is the bulk of that money going to come from?

It's probably (perhaps not with this product but say with the Grok version) inevitable, but is giving more help to oil, junk food, tobacco, dodgy politicians, and the odd millionaire who fancies themselves the next Il Duce what we want?

Oh hell yeah -- I had this idea but glad someone built it! Will definitely check it out

This looks awesome. I've used GPT to do something similar but having a platform like this would be very powerful. Congrats on the launch! Very excited to try it out soon

Thank you so much! Let us know what you think!! Free to try at app.societies.io :)

Does it consume your website to test with? Run js and use layout and positioning into account? What about local websites under development?

Recommend you don’t use loom had errors signing up to comment w google. Not wasting time on that. It was a cool demo and product, but you did gloss over, well skip, what the determining factors are that decide whether a user decides to interact with the post or not. Aka the secret sauce shouldn’t be secret. How is that determined what are the factors. Seems to be the most important part. Be clear on that and this product could have cool use. Congrats!

Thank you! Good tips haha, we've made a more polished launch video but thought to do a more dedicated loom for HN - feel free to check it out here: https://youtu.be/k6uo8PAn2vY

[flagged]

"Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something."

https://news.ycombinator.com/newsguidelines.html

Thankfully humans do tend to hallucinate haha

Congrats on launch. Impressive demo. Keep shipping!

Thank you so much!

Honestly sounds extremely dystopian. Thankfully I dropped social media a long time ago, but imagining some company creating a digital version of me and my friends so they can create hyper-focused advertisements to manipulate me into engaging with something is beyond gross.

Very sorry that it came across that way for you! We built Societies in the hope that the better people can understand each other, the better we can innovate solutions that meet people’s wants and needs - much of bad policymaking and bad product designs came from not being able to foresee the unintended consequences, and we hope that one day we can help flag those risks in an artificial society first, before taking the bet that could impact real people in real life :)

I am not going to enumerate all the ways this can be abused, but what controls do you have in place?

Every website you visit, every purchase you make online, every time you open safari on mobile, chrome on android, you’re being tracked. You don’t have to have social media anymore for a persona to be built for you.

yes, that is even more dystopian. If America was run by non blood sucking vampires (both parties) we would start heavily taxing and incentivizing outcomes now before society completely collapses.

That ship has sailed already… the only thing we can do now is start engineering Internet alternatives or better security. Crypto is broken when you have data centers full of H100s, quantum chips, and all Internet traffic routing through northern virginia.

"our biggest challenges are accuracy" lol

indeed - human societies are very complex haha. we have managed to predict how social media react to messages at a 80%+ accuracy, but still early days in making the simulation accurate across all contexts and all populations :)

best of luck!! great idea. would love it see how it executes.

thank you!! feel free to try it out by visiting app.societies.io on your computer :)

[dead]