GoKawiil
Tech News
New analysis finds that revenue per worker and profit margins are rising only among the Mag 7 tech giants, while remaining flat across the rest of the S&P 500. Productivity growth also shows no measurable lift from AI adoption, and non-tech corporations are reportedly treating AI token spending as a cost burden rather than a growth driver.
A discussion thread on Hacker News raises the question of why leading AI labs like OpenAI and Anthropic primarily sell model access as tokens rather than using their own frontier models to build and launch consumer or enterprise products directly. The poster argues that as token prices fall due to commodification, more profit potential may lie in building applications on top of models rather than selling raw model access.
Cysic Foundation identified an unauthorized mint of 10 million CYS tokens on BNB Smart Chain at 12:31 UTC on Oct. 17, 2026, and blacklisted the recipient address 76 minutes later. By then only about 2.8 million tokens remained at that address, with the other 7.2 million already transferred elsewhere and still being tracked. The foundation says preliminary findings point to a compromised database and flaws in bridge processing logic, and the bridge remains offline with no restart date.
A developer-focused blog post contends that relying on an AI provider's built-in analytics dashboard gives users little real insight into token usage or request behavior. Instead, the author recommends running a self-hosted LLM proxy that logs which API key called which model, from which client, and when, across multiple providers and models from one endpoint.
Neuralwatt CEO Chad Gibson analyzed a month of cloud traffic and found that different agent harnesses produce wildly different prompt-cache hit rates, from 1.6% to 11.1% uncached input tokens. The study found that a cache miss roughly doubles a request's energy use, and that deep sessions past 128k tokens with major prompt re-reads can cost as much compute as roughly 600 normal turns at that depth.
A technology retrospective recounts the history of Token Ring, the IBM-backed networking standard introduced on October 15, 1985, which competed with Ethernet through the 1980s and 1990s. The piece explains that while Token Ring avoided collisions by passing a single token between systems and handled network saturation well, it required all cards to run at the same speed and was ultimately outcompeted by cheaper, open-standard Ethernet equipment.
A developer has released Sygal, a tool that runs locally to convert images, scans, PDFs, web pages and YouTube videos into Markdown text. The creator says this conversion can cut token usage for AI models by roughly 2 to 7 times compared to feeding in raw images or markup, with YouTube links converted into clean transcripts.
Autheo's THEO token had been reviewed by Halborn, CertiK and Avelar Labs, but on October 13, 2026 a suspected bridge exploit allowed 19 million unauthorized tokens onto Base. THEO's price dropped from about $0.08 to around $0.006 within a day, and the attacker reportedly swapped the minted tokens for roughly 700.8 ETH, worth about $1.76 million, though that figure hasn't been confirmed on-chain. Autheo has told users to avoid buying, selling or bridging THEO while it investigates.
Docker disclosed a critical flaw in Docker Hub where converting a personal account into an organization could reassign that account's personal access token to a named owner without rotating the token or requiring the target's consent. An attacker could exploit this to authenticate as the target user, gaining the ability to pull, push, or delete images in their repositories, depending on the token's scope and permissions. The bug has existed since August 26, 2024, though it did not grant web access or allow changes to passwords, email, or two-factor settings.
A technical write-up describes how constrained decoding can turn a standard language model into a fast 'decision model' that picks from a fixed set of options (A through E) in a single forward pass, rather than generating output token by token as with structured JSON output. The piece includes example code using Qwen/Qwen3-1.7B to mask the vocabulary down to allowed answer tokens and select the highest-probability one.
A new open-source benchmark called the Telegraph Test shows that instructing large language models to answer in 'cablese' — the clipped, article-free style once used by telegraph operators — reduces billed output tokens by 40-49% on several models' own API meters, while preserving or even slightly improving factual recall. The effect held across four model families tested with roughly 1,300 questions over 50 passages, with one exception: gpt-5-mini's reasoning overhead made cablese prompting roughly double its cost instead of cutting it.
Futurum Group CEO Daniel Newman cited an Andreessen Horowitz chart of OpenRouter data showing AI agents used 7.3 trillion tokens versus 1.4 trillion for humans as of August, six months after agent usage overtook human usage. OpenRouter says over 85% of agent tokens come from cached prompts, and since February agent token usage has risen 14x compared with 2.8x growth in human usage.