AI Tokenmaxxing and Hypomania | Hacker News Reader

AI Tokenmaxxing and Hypomania

Been thinking about the whole ‘tokenmaxxing’ thing going on within companies. Couldn’t shake the feeling that it has all the symptoms of a mass hypomania. Now that ideas are churned, validated, and even built at the speed of thought, it feels there’s a huge surplus of energy that keeps competing in a rapidly changing tech landscape and fuelling this towards a full blown mania. It’s deeply alarming to even think about the downstream effects of the whole collective cementing a new baseline, forcing a cascade effect on society into a severe dissociation.

I'd love to understand HN folks thoughts here. Do you folks notice this pattern inside your orgs?

View on HN

I’ll just say that I my last meeting with my boss the only thing he cared about was that I was using AI, and how could I use AI more.

Then yesterday we hit our company AI quota for the month, so the whole company has to wait 2 weeks for it to reset. Entire teams are claiming they’re blocked. If it’s true we’re being measured against our AI usage, will work without AI during this time ever matter?

It also seems like a few tokenmaxxers could overrun the quota for the whole company. If the quota remains the same, I’m predicting we’ll see a race to use as much AI as possible after each reset, so people can get what they want out of it before it’s gone again. Potentially meaning more and more time each month without the tools for a majority of the organization. Time will tell on that one. It will be interesting to see what the company really values more, AI use or the money it costs them.

Blocked? That's atrophy, right?

And yeah, token usage as a KPI is a trend I notice across orgs too. Mostly in the 'hypomania' stage where the company is blind to the ROI they're getting out of AI. In saner companies, at least short-term, the human arbitrage should get realised during cooldowns like this. But then again, we're just a few cycles away from better quantization and efficient inference engines. Meaning, relaxed token limits. Also, there's local models getting on par with cloud models. At a point, the quota friction will disappear. The measure will get normalised! That's a hard thing to imagine tbh.

> Blocked? That's atrophy, right?

Even before AI I noticed this trend. The second someone gets a new tool, they seem helpless without it. The skill atrophy seems almost instant. Even if they have the skill, many people seem unwilling or unable to use it.

> Meaning, relaxed token limits.

I wonder how this will play out. It seems like AI companies are incentivized to make tokens seem rare and valuable, while also encouraging heavy use and tokenmaxxing, to boost revenue. It seems like that will only be challenged when a price war between frontier models, or local models being good enough and cheap/easy enough to make the frontier models seem redundant.

The M2 money supply has increased by roughly 48% since 2020. There are probably social consequences to this (as it would be strange if there weren't). Perhaps we are feeling the need to try every possible idea to see if that is the thing that wins the money (which is new, and does not obey the old rules).

Thanks for this lens. This means, we are indeed structurally entering a state of dissociation with surplus capital and AI-enabled 'bandwidth' being the catalysts. The only friction is compute.

Reply on news.ycombinator.com