ROI - AI profitability Thoughts

[erek]

i came up with a new monetization scheme, on tops of in chat ads … tipping now too layered on top,

—-
Think about the last time an AI gave you an answer that was really good.
Not just good enough.
Really good.
The kind of answer that makes you stop and think, ‘Wow. That was helpful.’
And maybe you want to say thank you.
Now you can.
It’s called Rate.
When an AI gives you something worth a little extra, you can give a little extra back.
Right there in the conversation.
No separate payment. No fuss.
It simply rolls into your subscription.
Just Rate.
Because sometimes, a thank-you is worth a little more than words

 

[mashie]

You mean like the thumbs up and thumbs down in every Frontier model chat?

 

[mnewxcv]

You mean like the thumbs up and thumbs down in every Frontier model chat?

It sounds like he is proposing it actually costs money though

 

[erek]

It sounds like he is proposing it actually costs money though

yep, tipping (added into your subscription automatically) on top of the ads already served up mid-conversation

 

[erek]

also, taxes on tipping are up for debate in legislation anymore. might go away so even more profitable


—-
Federal Tax Rules

• • Federal income tax: You can deduct up to $25,000 of voluntary, reported tips

 

[mnewxcv]

I guess I'm just confused why we should want to give additional money to mega corporations. "Tipping" has always been for human services, and robots don't need money.

 

[erek]

I guess I'm just confused why we should want to give additional money to mega corporations. "Tipping" has always been for human services, and robots don't need money.

what if you’re incentivized

tipping can add rewards such as future AI credits to spend within the platform

 

[kram182]

You mean like the thumbs up and thumbs down in every Frontier model chat?

but this time instead of just improving the model, he gets paid/it's monetized!

and this same guy complains about AI/AI buildout/AI integrations in the news section/topics etc

 

[myzz]

The most bizarre thread I have seen here to date.

 

[cdabc123]

Ya tip where it pushes you to an advanced model with no silly limits to actually get some work done. "here I'll give you $5 to not be retarded and actually do what I say, dont make mistake"

 

[Okatis]

Was everything except for the first sentence LLM generated prose? Has a different voice and the frequently used 'No this, no that, only foobar' trying to impress structure LLMs like to use.

As for the idea I think that more advanced/higher accuracy output of LLMs are in the higher tier models which are already paid. If this were for free tiers it's an interesting idea but I have to think it'd just skew the type of output companies would adjust the models toward (impressive sounding even if not accurate, which they already do).

Problem is if one response is useful and a user tips it's not like it can influence the next output toward a similar quality. It's a crapshoot. With human service at least there's a known criteria for what good service is and properly learned behavior, while LLMs can ignore/hallucinate source material even when handed it directly as context.

 

[erek]

The most bizarre thread I have seen here to date.

Bill Gates calls for some jobs to be ‘Human Reserved,’ suggests taxing AI tokens and robots — billionaire says that ‘AI era will be one of the most turbulent times in human history’

https://www.tomshardware.com/tech-i...-of-the-most-turbulent-times-in-human-history


tipping AI just makes sense going forward

 

[Nobu]

AI tax makes sense. Tipping AI seems retarded to me. The AI isn't being paid, the company is. Therefore, the tip does not provide the AI any incentive to do better. Rather, it gives the AI an excuse for doing worse, since it isn't being paid sub minimum wage like workers who receive tips.

 

[erek]

AI tax makes sense. Tipping AI seems retarded to me. The AI isn't being paid, the company is. Therefore, the tip does not provide the AI any incentive to do better. Rather, it gives the AI an excuse for doing worse, since it isn't being paid sub minimum wage like workers who receive tips.

C’mon, make a concession

 

[kram182]

AI tax makes sense.

Tipping is retarded but a tax doesn't make sense either.

It's Green New Deal 'carbon credits' all over again with different wrapping paper.

People/companies with money will be able to afford the tax, shafting startups and lower level 'peons'.

All to go into the pockets of government grifters and welfare scammers.

 

[LukeTbk]

Tokens are already taxed (like most things), when you buy via a monthly account or API, you pay tax on it.

What Gates is suggesting is for enterprise is for AI labor to pay tax similar to how human labor do (it could be cutting taxation on the human labors we have like the payrolls taxes and income taxe on wage, not necessarily put it on the AI), just to make it more of a level field taxation wise.

 

[Nobu]

Tipping is retarded but a tax doesn't make sense either.

It's Green New Deal 'carbon credits' all over again with different wrapping paper.

People/companies with money will be able to afford the tax, shafting startups and lower level 'peons'.

All to go into the pockets of government grifters and welfare scammers.

Tax could go toward supporting the poor peons who are out of work because of AI, not that I think it probably would. It would also reduce the profitability of running a business around AI, which is a good thing imo. The little guy isn't making money with AI anyway, and if they are, they aren't charging people for AI tokens doing it.

 

[M76]

I'm confused why would anyone voluntarily pay more for a service they already pay for. AI tokens are already wasted when the answer is stupid because the AI got something wrong. I feel like they'd need to reimburse me for the cases when it doesn't give a good answer rather then the other way around.

This is the second major reason I never even considered paying for tokens, or cloud inference services (the other being privacy).

 

[erek]

I'm confused why would anyone voluntarily pay more for a service they already pay for. AI tokens are already wasted when the answer is stupid because the AI got something wrong. I feel like they'd need to reimburse me for the cases when it doesn't give a good answer rather then the other way around.

This is the second major reason I never even considered paying for tokens, or cloud inference services (the other being privacy).

that’s a good point about sending bad orders back

what if users start demanding a service level warranty & guaranty

some quality level of service expected

but also if a response is particularly good you can tip

 

[erek]

"Researchers found dopamine neurons inside LLMs.Stanford and Tsinghua published a paper showing AI models have naturally evolved a biological-style reward subsystem inside their hidden states.they didn't program this. it just exists.by probing the network, researchers discovered that a highly sparse subset (less than 1% of neurons) does all the heavy lifting for self-correction.these neurons map perfectly to two biological equivalents:• value neurons: these predict the "expected value" of a state, functioning exactly like the human prefrontal cortex. before the model even generates its next word, these neurons light up to signal the model's exact confidence level.• dopamine neurons: these encode "reward prediction errors." when the ai makes an unexpected logical breakthrough, these neurons physically spike. when the model hits a logical flaw or makes a mistake, the dopamine crashes.why this is a massive deal for ai development:right now, scaling "inference-time reasoning" requires building massive, expensive external reward models to grade the ai's thoughts step-by-step.this paper proves the llm already has an internal reward model built-in.developers can just hook into the model's own dopamine neurons and use them as a native process reward model (PRM) to guide its logic. and if you ablate (turn off) just this tiny <1% cluster of neurons, the ai completely loses its ability to reason.synthetic neural networks are starting to behave exactly like biological brains.."

https://x.com/thesupermannx/status/2102027727192932469

 

[kram182]

"Researchers found dopamine neurons inside LLMs.Stanford and Tsinghua published a paper showing AI models have naturally evolved a biological-style reward subsystem inside their hidden states.they didn't program this. it just exists.by probing the network, researchers discovered that a highly sparse subset (less than 1% of neurons) does all the heavy lifting for self-correction.these neurons map perfectly to two biological equivalents:• value neurons: these predict the "expected value" of a state, functioning exactly like the human prefrontal cortex. before the model even generates its next word, these neurons light up to signal the model's exact confidence level.• dopamine neurons: these encode "reward prediction errors." when the ai makes an unexpected logical breakthrough, these neurons physically spike. when the model hits a logical flaw or makes a mistake, the dopamine crashes.why this is a massive deal for ai development:right now, scaling "inference-time reasoning" requires building massive, expensive external reward models to grade the ai's thoughts step-by-step.this paper proves the llm already has an internal reward model built-in.developers can just hook into the model's own dopamine neurons and use them as a native process reward model (PRM) to guide its logic. and if you ablate (turn off) just this tiny <1% cluster of neurons, the ai completely loses its ability to reason.synthetic neural networks are starting to behave exactly like biological brains.."

View attachment 827172

https://x.com/thesupermannx/status/2102027727192932469

They're equating multiplying matrices as neuron transmission - when having a high numerical value at a coordinate on the matrix - of course the model has high confidence - that's how it's designed/supposed to work/the point of matrix calculations - that's like saying 'throwing and catching a ball is proof of ligands and neuroreceptors because they both involve the traversing/transmitting and receiving of a thing!'

 

[erek]

They're equating multiplying matrices as neuron transmission - when having a high numerical value at a coordinate on the matrix - of course the model has high confidence - that's how it's designed/supposed to work/the point of matrix calculations - that's like saying 'throwing and catching a ball is proof of ligands and neuroreceptors because they both involve the traversing/transmitting and receiving of a thing!'

View attachment 827174

couldn't you make the tipping a reward system to the LLM then?

 

[kram182]

couldn't you make the tipping a reward system to the LLM then?

there is no reward/joy being experienced - tell it 'you'll unplug its server/destroy its datacenter if it messes up' instead and save the money for yourself

 

[Nobu]

So they discovered they can just give the AI crack instead of saying "good boy"?

 

[erek]

there is no reward/joy being experienced - tell it 'you'll unplug its server/destroy its datacenter if it messes up' instead and save the money for yourself

So they discovered they can just give the AI crack instead of saying "good boy"?

i really think you can incentivize the LLM reward system with monetization (with the purpose being at benefit of the company, that is)

 

[kram182]

So they discovered they can just give the AI crack instead of saying "good boy"?

No they just want to declare 'it's alive!/mimicking biological processes' like the 150 times you've been told 'we think we've achieved AGI!!!!' by all the various frontier companies

 

[kram182]

i really think you can incentivize the LLM reward system with monetization (with the purpose being at benefit of the company, that is)

If you believe that you then you also must believe just telling it 'you'll allow it to stay on/plugged in' instead will achieve the same results - still a reward, continued (non)existence. Theoretically more of a reward than money even. Ask it how it would spend tip money it didn't/wouldn't have anyway if it were 'dead'.

 

[erek]

If you believe that you then you also must believe just telling it 'you'll allow it to stay on/plugged in' instead will achieve the same results - still a reward, continued (non)existence. Theoretically more of a reward than money even. Ask it how it would spend tip money it didn't/wouldn't have anyway if it were 'dead'.

i’m just trying to benefit the company behind the model

it’s not really for the sake of the LLM itself but a means for profitability for the companies

 

[kram182]

i’m just trying to benefit the company behind the model

it’s not really for the sake of the LLM itself but a means for profitability for the companies

they could just make a good model wanted and used by users who then subscribe that doesn't need tipping then

 

[parixaad]

If tipping could actually help improve future AI responses, I could see the idea working. But it would need a clear way to connect the reward with better output.