In the author’s comparison of 200 biopsy-confirmed skin lesions, a fine-tuned version of 2015’s Inception v3 correctly identified the diagnosis from six options in 62% of cases; two LLMs managed 42%. The sample was small, included 30 melanomas, and used a model that is not state of the art.
On 200 biopsy-confirmed skin lesions, fine-tuned Inception v3 achieved higher diagnostic accuracy than two LLMs.
02CHIPS & INFRASTRUCTURE
Cerebras explains how it spread a 3-trillion-parameter model across 59 wafers
In an example using 4-bit weight quantization, an assumed 256,000-token request, and an 8-bit KV cache, Cerebras calculates that Kimi K3’s 69 linear-attention and 24 dense layers require 59 WSE-3T wafers. The post is an illustrative model-placement calculation, not a production-performance benchmark.
Cerebras describes in a technical post how it runs Kimi K3 by distributing its layers across 59 WSE-3T wafers.
03INDUSTRY
Apollo economist: AI returns are not evident outside tech companies
Apollo Global Management chief economist Torsten Slok says profit margins and revenue per employee have remained flat at S&P 500 companies outside the technology sector, with no evidence that AI has boosted productivity. This is Slok’s assessment; the text gives neither a measurement period nor a detailed methodology.
Apollo chief economist Torsten Slok argues there is no evidence that AI has increased profit margins or revenue per employee at large companies outside the technology sector.
04PRODUCTS
Google’s AI search is heightening publishers’ traffic concerns
Google’s AI Overviews and AI Mode put answers front and center in search results, raising publishers’ concerns about traffic losses. DCN says Google traffic to 19 of its members fell 1–25% in May–June 2025; Google, meanwhile, says click volume has remained relatively stable.
Traffic losses associated with Google’s AI Overviews and AI Mode have pushed publishers toward social media and subscriptions; Google says click volume is relatively stable.
05RESEARCH
AI’s role in US creative-sector job losses remains unclear
The article says US creative industries lost more than 200,000 jobs over the past four years, with losses concentrated especially in film and sound recording. AI’s share cannot be measured; media consolidation, outsourcing, and social media may also have played a role.
An article reports more than 200,000 job losses in US creative industries over four years and stresses how hard it is to separate AI’s impact from other pressures.
06POLICY & LAW
Nikon withdrew contest winner’s award over AI use
Nikon revoked Ning Xu’s first-place award in its Small World in Motion contest after finding a rules violation. Xu said he used AI during image processing, not to create the video or cell movements. Scientists questioned image anomalies; Nikon said it would review future contest rules and judging.
Ning Xu, who won first place with a microscopy video, lost the award after Nikon concluded that the contest rules had been violated.
07RESEARCH
Shared notes boosted coding performance
The study presents a training-free method that coordinates model instances using plans and notes in a shared file system. On LiveCodeBench’s 100 hardest problems, open-weight Qwen3.8-27B rose from 66.8% to 92.4% accuracy. However, the gains did not appear across all models.
A study reports that coordinating model instances through shared files containing plans and notes can improve some models’ LiveCodeBench performance.
08DEVELOPER TOOLS
Pruning and resets tackle long-context problems in agents
The speaker argues that context engineering is shifting from large context windows to code-based “harnesses” that manage agents’ tools and history. Long contexts can cause distraction and fixation on goals; pruning, summarizing, and resetting are used as remedies. The speaker also predicts that accumulated personal agent memories may conflict.
A presentation argues that managing context in AI agents is shifting from prompt writing to code-based systems that manage tools and history.
✓
That’s it for today.The rest of your time is yours.
Your daily technology digest, every morning in English. Free. Unsubscribe anytime.
Sources, in the open.
Stories are selected from Hacker News; accessible original articles are summarized and translated with AI. Summaries may contain errors; check the source beneath each story for details. Preparation starts daily at 06:55 Türkiye time (UTC+3), targeting publication before 07:00.
This edition covers articles shared on Hacker News on the date shown above.
Now on the App Store for iPhone
Your daily digest. Now in your pocket.
Read or listen to a new edition every morning with Kısa for iPhone, save stories for later and keep your streak.