Hacker News Digest

Thursday, June 25, 2026

In This Issue

  • Hacker News
  • OpenAI unveils its first custom chip, built by Broadcom
  • Founding a company in Germany: €9600, 152 days and I still can't send an invoice
  • Jerry's Map
  • John Carmack on the mistakes around Quake that ruined id software
  • Anthropic Accuses Alibaba of 'Illicitly' Accessing AI Models
  • RubyLLM: A single, beautiful Ruby framework for all major AI providers
  • For Most of the World, Open-Source AI Is the Only Way Forward
  • Computer use in Gemini 3.5 Flash
  • Reid Hoffman says SpaceX 'not an AI company', xAI 'complete train wreck'

Zipper Data Brief

June 25, 2026
Your daily digest of the best from Hacker News

Top 6 Trending

#1
720 points · jamdesk · comments
OpenAI's custom inference chip, made by Broadcom and arriving in 2026, promises 50% cost savings over GPUs, marking a strategic move toward hardware self-sufficiency—though commenters question the timeline, Broadcom's reliability as a partner, and whether such specialized chips will become obsolete as AI efficiency improvements accelerate.
#2
582 points · earcar · comments
# Summary The discussion reveals that while German company formation bureaucracy is genuinely complex and inefficient, the author's experience was worsened by choosing an unnecessarily complicated corporate structure (UG & Co KG) and relying heavily on lawyers—simpler alternatives like a basic UG or GmbH would have been faster and cheaper. Commenters widely agree Germany's startup environment is burdensome compared to other countries like Estonia or the Netherlands, though some argue the issues are overstated by those who don't follow standard procedures.
#3
574 points · turtleyacht · comments
# Summary The discussion celebrates Jerry's Map, a 50-year solo project where an artist incrementally draws and updates map tiles following a card-based rule system, with commenters drawn to its meditative quality, outsider art appeal, and the poignant fact that the artist only sees the complete map years apart. Many share nostalgia for similar childhood mapping projects and appreciation for the systematic, obsessive nature of the work.
#4
538 points · shadowtree · comments
# Summary John Carmack reflects on how pushing the team too hard during Quake's development burned out id Software's talent and fractured the company, though most commenters agree Quake's technical and cultural impact justified the cost, while debating whether the company's later decline was inevitable or preventable.
#5
486 points · htrp · comments
# Summary The discussion centers on the accusation that Anthropic is hypocritical for complaining about model distillation by Chinese competitors, given that Anthropic itself trained Claude on vast amounts of copyrighted internet content without permission. Most commenters argue this represents "thieves complaining about thieves" and contend that distillation is an inevitable, unprotectable part of AI development that Anthropic cannot legitimately prevent.
#6
396 points · doener · comments
# Summary RubyLLM is widely praised for its elegant API design and ease of use in production, though users note friction points around maintainer responsiveness, provider-specific feature limitations (like prompt caching inconsistencies), and observability/debugging challenges. The framework shows strong adoption in the Ruby community despite broader concerns about Ruby's declining popularity.

AI / Machine Learning

217 points · CrankyBear · comments
# Summary The discussion centers on whether open-source AI is feasible and desirable, with commenters debating the definition of "open-source AI," questioning whether ordinary people can afford to run capable models, and arguing that true openness requires not just open weights but also open hardware and computational accessibility to counter potential monopolistic control by a few corporations and nation-states.
227 points · swolpers · comments
# Summary Users report significant frustration with Gemini 3.5 Flash's limitations in practical tasks—including poor instruction-following, dangerous automated actions (like running destructive git commands), excessive safety guardrails, and lack of competitive features like proper coding tool integration—though some praise its speed and cost-efficiency compared to competitors.
231 points · 1vuio0pswjnm7 · comments
# Summary The discussion largely dismisses Reid Hoffman's criticism of xAI as self-interested commentary from an OpenAI investor with a history of conflicts with Musk, though commenters debate whether SpaceX qualifies as an AI company based on its Cursor acquisition and datacenter rentals, and whether xAI's models are actually competitive despite execution challenges.
388 points · mattnewton · comments
Krea released open-weight text-to-image models (Krea 2 Turbo and Raw) with detailed technical documentation, positioning them competitively with existing models, though some debate whether the restrictive license (under $1M revenue threshold) qualifies as truly "open source."
How to find AI-conservative companies to work for?
19 points · tossitawayplz · comments
# Summary People seeking AI-conservative jobs should look at regulated industries or non-tech hubs where adoption is slower, though AI is still present everywhere in software engineering. However, one commenter argues that resisting AI is ultimately futile and counterproductive—the technology is a valuable tool that, when used properly with good architecture and documentation, enables individual engineers to accomplish more.

Startups / Business

47 points · zayd7861 · comments
A 22-year-old solo developer built Wealtii, an on-chain index fund platform with 315k lines of code using FastAPI/React/Postgres, batching user orders to cut gas costs by 80% while maintaining non-custodial security through multi-sig wallets.
60 points · gsky · comments
I cannot provide a summary as the only comment shown is a duplicate flag pointing to another discussion thread. No substantive discussion content is presented to summarize.
67 points · colesantiago · comments
# Summary The discussion about Qualcomm's acquisition of Modular was moved to a different HackerNews thread that was currently on the front page, so no substantive comments are available in this thread.
22 points · wslh · comments
The discussion is sparse, with a single skeptical comment questioning the financial feasibility of building a hard tech company despite the allegedly favorable conditions mentioned by the former Meta CTO.
14 points · isomorphic_duck · comments
Google is losing AI researchers to Anthropic, with speculation that departures may be driven by Gemini's underperformance and concerns about Google's proprietary hardware (TPUs) compared to industry-standard alternatives that Anthropic likely uses.

More Stories (34)

164 points · EthanFantl · comments
64 points · toomuchtodo · comments
320 points · panza · comments
267 points · felixdoerp · comments
109 points · Imustaskforhelp · comments
Ask HN: Where is our profession (programmer) going?
63 points · syntaxbush · comments
168 points · phreddypharkus · comments
125 points · dimastopel · comments
68 points · harsehaj · comments
Tell HN: I never bought anything from clicking on a paid ad
22 points · julienreszka · comments
193 points · dakrone · comments
Ask HN: What home printer do you use/recommend?
17 points · niyazpk · comments
Created by Zipper Data Co.  · 2026-06-25 12:01 UTC  · Unsubscribe

Get digests like this delivered to your inbox every morning.

Subscribe Free