683 points · tosh · comments

# Summary The discussion reveals that Claude's system prompts have grown from ~300 to 3000+ words and contain extensive instructions on safety, behavior, and policy handling—with commenters debating whether this verbosity is necessary, noting contradictions, questioning why such details aren't baked into model weights, and expressing frustration that verbose instructions may actually degrade performance rather than improve it.

667 points · pentagrama · comments

# Summary While Firefox's new native adblocker is a welcome addition, commenters note it's late to the game compared to competitors like Brave and Safari extensions, lacks key features (like search result ad blocking), and still doesn't match the customization of uBlock Origin or support iOS's Content Blocker API.

560 points · bilsbie · comments

# Summary Qwen 3.8 27B delivers impressive reasoning capabilities on consumer hardware, but its default "overthinking" behavior makes it token-inefficient and slow; users are finding workarounds by adjusting reasoning effort settings or using custom forks to balance quality with practical performance trade-offs.

547 points · stagas · comments

# Summary Cloudflare automatically injects analytics JavaScript into free-tier websites by default, which users can disable but is difficult to find. This practice is only possible when using Cloudflare as a reverse proxy (not just DNS), raising concerns about their growing use of dark patterns and silent feature injection.

542 points · Narishma · comments

# Summary The discussion reveals a fundamental disagreement about RISC-V's value: while the original critic argues it has design flaws and fragmentation issues, this response contends RISC-V's open licensing and cost benefits for embedded systems in developing countries outweigh its technical shortcomings—though commenters note the article's shipping cost argument is logically inconsistent and overlooks how freight forwarding actually works in these regions.

406 points · Alifatisk · comments

# Summary The discussion explores "tortured phrases" like "kidney disappointment" (instead of "kidney failure") found in academic papers, which result from plagiarism-avoidance tools that automatically replace words with awkward synonyms, a problem predating modern AI that has plagued scholarly publishing since at least 2021.

121 points · healsdata · comments

# Summary Young people's distrust of AI CEOs reflects broader skepticism toward tech leadership and wealth inequality rather than AI itself—a sentiment commenters argue is either reasonable skepticism about corporate motives or media-manufactured outrage depending on one's perspective.

318 points · hruvhwe · comments

# Summary The discussion is heavily critical of an AI-generated article claiming models should separate reasoning from knowledge. Commenters point out the article contains factual errors, oversimplifies the inseparability of knowledge and reasoning, and ignores that RAG/tool-use doesn't eliminate hallucinations—while also noting the irony that an article about AI dumbness was itself AI-generated.

35 points · herbertl · comments

# Summary The discussion centers on the feasibility and challenges of creating "Humazon," a human-authored book platform to counter AI-generated content. While commenters acknowledge AI detection is an ongoing arms race with potential solutions like blockchain authentication or proof-of-work verification, they note that implementing reliable human-verification at scale would be resource-intensive and legally fraught.

301 points · mlenhard · comments

# Summary A vibrant gray market has emerged for reselling AI API credits at steep discounts (up to 98%), driven by account creation abuse, stolen credentials, and YC startup perks—raising questions about fraud, data security risks, model distillation opportunities, and whether AI companies' unsustainable economics are fueling the black market.

19 points · ryanmerket · comments

# Summary Kimi Work automatically attaches the last 5 user sessions to feedback reports without explicit user consent or notification, which raises privacy concerns—unlike Claude, which clearly discloses that feedback submission shares conversation data.

384 points · zacharyozer · comments

# Summary The discussion reflects mixed reactions to Stripe's $7B acquisition of OpenRouter: supporters see strategic value in Stripe leveraging its API/payment expertise to become an LLM infrastructure middleman and data collector, while skeptics question whether a relatively simple proxy service justifies the valuation and worry about future censorship, privacy implications, and service quality under corporate ownership.

14 points · sbulaev · comments

The acquisition positions Stripe to control AI model access and usage metering, raising concerns about whether it will remain neutral or favor certain providers.

13 points · ishener · comments

No comments available.

43 points · root-parent · comments

# Summary Commenters are skeptical of Anthropic's $190-200B 2028 revenue forecast, questioning whether the valuations are realistic given competition (particularly from cheaper alternatives like DeepSeek), lack of profitability details, and the enormous market growth assumptions required to justify such projections.

214 points · root-parent · comments

The discussion questions Nvidia's scaling back of financing guarantees for OpenAI's Ohio data center, with commenters expressing skepticism about unsustainable AI financing practices, inflated valuations, and whether Nvidia's profitability can withstand the risks if the deal fails or open-source models disrupt the market.

239 points · walrus01 · comments

156 points · brandonb · comments

130 points · Jimmc414 · comments

195 points · ronbenton · comments

74 points · Jimmc414 · comments

127 points · thunderbong · comments

102 points · antipurist · comments

75 points · NordStreamYacht · comments

188 points · toomuchtodo · comments

52 points · _delirium · comments

82 points · karakoram · comments

32 points · petethomas · comments

49 points · rib3ye · comments

44 points · Jimmc414 · comments

27 points · _____k · comments

23 points · buffer_overlord · comments

22 points · anigbrowl · comments

22 points · Alien1Being · comments

20 points · ksec · comments

27 points · Animats · comments

19 points · yusufaytas · comments

18 points · bradley_taunt · comments

29 points · nabeards · comments

16 points · rbanffy · comments

364 points · ropbear · comments

96 points · erickhill · comments

17 points · gpi · comments

23 points · vlowrian · comments

95 points · spdegabrielle · comments

94 points · matt_d · comments

13 points · freakynit · comments

13 points · Jimmc414 · comments

104 points · the-mitr · comments

31 points · AnodicElegy · comments