aditya_rs

Thank you for reading it and providing notes.

> If you are saying sentience is 'larger' or different to intelligence then I'd argue you are defining it,

I suppose it's more of a character definition rather the definition of thing itself, it makes sense? Like how you would say Green is not Red or darkness is absence of light.

> No big name in AI academia or philosophy of mind seems to have made a strong claim about sentience in AI that we currently have. A lot of laymen have made claims, without defining sentience, or consciousness

I think for me this shouldn't prevent us from making qualitative delineation between these things without having to resort to waiting for a full definition for these things to be carved out. As to why should we even bother making this distinction, I think it's to have some clarity on things like morality when talking about these machines which can replicate with surprising fidelity the only means we have of expressing our inner world.

> This exists. It is called Cognitive Studies

Thank you for pointing me to this direction. I still feel there is something lacking? I can't put my finger on what it is, like just intelligence has to be seen independently as a mechanical field instead of clubbing it with other things. Though it might be a recency bias of having written this article.

> it's an emergent property based on size of the network of connections and underlying training data (be that human sensory input or machine vectors). Trim an LLM down and you won't see 'intelligence' disappear at a binary boundary, it'll degrade gradually to various different points of usability.

I thought about and continue to think about this. Regarding humans there is still open discussion on are babies conscious https://www.nytimes.com/2026/10/05/science/consciousness-bra...

Yep, that's what I think and wrote on my blog. My hope is that the inference providers should provide a managed service like it. It's also in their best interest to do so, because if they don't and these provider hosted agents become the norm, demand for independent inference would drop.

Possibly so for now. But do you think the providers are interested in making interoperability easy?

Even in the future if they are required by law to provide it, I wouldn't bet against the craftiness of the providers to invent some sort of network effect dark pattern to make it painful, if not outright impossible.

Just as an example, with Muse I can already see that the way they are thinking of making money is via taking a transaction cut, so it's not that hard to imagine that Meta can negotiate deals for txns that happens through Muse which won't be available elsewhere.

The series is also cancelled, so no point wasting time watching it either.

I feel like a lot of these always on agents tie users deeply into the platform. Unlike models that you can swap between with relative ease, with an agent because of the integrations to other platforms, work history and so on it would be harder to switch, since in effect they are essentially your computer on the cloud.

Tin foil hat version of me thinks that all the closed model companies want to desperately build an abstraction layer on top of the model, so that they can limit access to the model directly and build a locked down relationship with the user.

Other inference providers should counter this by providing their own version of standardized managed agents.

Coincidentally, I published a note on my blog just about this today https://aditya.rs/blog/2026/09/29/inference-providers-should...

Was just working on https://github.com/adityarsuryavamshi/Sniper which is a Manifest V3 extension which does dynamic user action execution without needing `userScripts` permissions.

I was mostly interested in solving a very specific problem for myself (basically poll a website behind auth + 2fa and react/notify to changes) and since I also had recently gotten into frontend decided to also try out building extensions.

I knew about Tampermonkey and such but had never used them, and I also wasn't sure if they would be long term viable with MV3 coming in. Moreover I also wanted to explore some way to keep options open in case chrome ended up being very heavy handed with extensions, in the process I discovered there are still a lot of things you can do if you really really wanted to bypass the current limitations.

In the end though I personally decided to use Violentmonkey instead of this, but this could be an alternative direction for others who don't/can't use dev mode for extensions.

Tried to answer a similar question here https://news.ycombinator.com/item?id=31245514

Indeed, after thinking about this, prediction markets was one of the things that came to my mind, this is something I hopefully will research a bit more on. I was primarily interested in the general pattern of using the negative outcomes to align incentives and also potentially use it actually solve the problem. Prediction markets are possibly a pretty good place to experiment this.

> Also, wouldn't this lead to weird incentives for the market participants? Like in an extreme case, murder?

Yep, this is the reason why I think we should continue to have a regulatory stopgap to provide a floor for acceptable behavior

> but I think it's too central to the whole idea to leave as a handwave

haha yeah, you're right I could have been more clear on why I thought there was a need for it.

One way to look at it (though might not be a strong argument) is even with the current regulatory framework, the incentives are not aligned. So I don't think regulation is a way to align incentives rather it acts as a floor of what should be acceptable behavior.

The reason why I think a regulatory stop gap is needed is, when a problem doesn't have enough momentum towards it, it's much easier to make a problem worse than to make it better and regulation acts as a valve by indicating the directional progress we want to make and prevent making it worse at least until there is sufficient momentum that progress from the problem is clearly much better than the alternative.

That said, I think for the problems I used as examples in the article the current mechanisms and the current framework we have are sufficient, and possibly no new regulation needs to be taken. (i.e, for the case of doctors, we already have pretty good mechanisms to prevent doctors from creating actual harm, for climate change we already have limits on carbon emissions, etc.)

> Someone needs to actually issue the futures, and I'm not sure who would do it; these 'futures' contracts seem more like straight bets than actual futures.

You're right I didn't mention specifics of on who would create these. A naive answer would have been "why not anyone?", but you bring up an interesting point on who would actually hold the bet for a long duration of time and be guaranteed for pay-offs.

I disagree with the problem of enforceability of the paying out in case someone looses a long position, since it's also a current problem with the standard futures for stocks/commodities, where the brokerage firm would be liable for paying up in case of defaulting on a margin call, so similar structure might be imagined for these cases.

With regards who might create these futures, the government seems like a good candidate, but really if we do have a solid clearinghouse with similar structure for handling margin calls etc. I don't see why not anyone can create these.

The regulatory concerns for implementing these is very valid, but I was primarily just interested in exploring the idea.

I'm assuming you're question is why would they buy the futures?

If they really believe that climate change is not real, then they would benefit from owning the futures which pay off when if the climate really doesn't warm more than 2°C in the next 3 decades.

Pay for performance is a perfectly valid way to solve these problem if the outcomes are clearly attributable for the service provided, but consider cases like strategy consulting where the outcome of hiring a strategy consultant is hard to attribute.

One of the things I was interested about is problems where there is possible deniability and attribution ambiguity, in these cases my thinking was it is better to scope and anchor to a larger issue that might occur and align incentive towards a more holistic outcome rather than specific service. The futures approach was a general way of doing it which is agnostic of the problem in hand (though the specific implementation of it would inevitably be tied to the outcome you want in the end).

The doctor example is a common example I keep using since it is very simple to explain, it is purely for exploratory purposes and I did note that it is quite possible not all doctors and consultants might do this. My point was not to deride anyone but rather to just explore an idea.

The enterprise expansion of 1Password seemed like a natural evolution for 1Password. If you build your entire product extensively focused on a single quality you'll naturally end up with something of a platform. I wrote more about it here specifically taking 1Password as an example https://aditya.rs/2022/02/08/qualities-and-business-models/#...

Congrats on the job offer!

If you don't mind sharing what was the level you were interviewing for and how long did the whole interview process take?

Thanks for the detailed answer!

I think in general I can see that startups are exploiting the interviewing process itself as an arbitrage to get smart people in. Large companies to some extent are very set on the interview "loop" and would go through the whole assortment of their test suite irrespective of who it is, maybe they can afford to and they may even have to at scale as a process discipline to avoid false positives. But this is a clear advantage for startups hiring since by their nature itself get lower inbound so they can customize their interviews to either make it more specific or remove certain rounds altogether depending on the signal they get from the candidate.

Congrats on the offer!

> e) Not really necessary

What do you mean? As in you didn't have a need for managing the time because you weren't really interviewing for a lot of companies or you didn't find the number of interviews too much of a burden on your time?

Congrats on the offer! Do you mind sharing the level you were interviewing for and how long did the whole interview process take?

Congratulations on the offers! Did you have not have any leet code kind of interviews, I should asked this in the question, but do you mind mentioning what level were you interviewing for and how long did your whole interviewing process last (i.e, from when you first started interviewing for a company to when you stopped)

Haha Nokia didn't even cross my mind when I was writing this. But when I googled connecting people it was one of the first page results, so I can understand the confusion.