747 points · Zeekaas · comments

# Summary Hackers exploited well-known vulnerabilities in outdated Danfoss refrigeration controllers that were left unpatched and directly exposed to the internet, affecting European supermarkets; the incident mirrors a similar attack on US military bases weeks earlier and highlights how cost-cutting by manufacturers and retailers creates critical security gaps in essential infrastructure.

604 points · prophittcorey · comments

# Summary The discussion explores building modern web applications in C using CGI, with commenters appreciating the simplicity and efficiency of the approach while debating tradeoffs like concurrency handling, security concerns, and whether simpler architectures still make sense versus complex modern stacks. Opinions are mixed on whether CGI with C is practical today versus alternatives like Go or embedded HTTP servers.

590 points · ilreb · comments

# Summary The discussion expresses skepticism about GLM-Legal's practical utility, citing concerns about hallucinations in novel cases, jurisdiction-specific reliability, privacy risks with Chinese servers, benchmark validity, and the fundamental need for qualified human oversight in legal work—though some commenters acknowledge LLMs could assist with routine, generic legal tasks.

574 points · whimblepop · comments

# Summary Users report that Codex CLI installs a persistent, self-updating background daemon by default that causes numerous problems including disk bloat (12GB+), permission bugs, and reliability issues, though it can be disabled via config flags or `--no-daemon`. The broader criticism highlights poor software quality, lack of adherence to standard package management practices, and frustration with the trend of AI tools implementing their own update mechanisms rather than using OS package managers.

452 points · layer8 · comments

# Summary The discussion is deeply divided on whether AI truly requires skill to use well: skeptics argue that rapidly improving models are making previous skills obsolete and that most people will simply outsource thinking rather than develop genuine expertise, while proponents contend that meaningful skill gaps persist (knowing what to ask, maintaining project complexity, catching errors) and will likely continue despite AI improvements—though some worry these skills themselves may become irrelevant as models advance.

362 points · djfergus · comments

# Summary Developers have discovered how to repurpose WiFi 7 router chipsets as software-defined radios (SDRs) capable of receiving at 320 megasamples per second—an impressive capability that's cheaper and more accessible than traditional SDR hardware, with similar efforts underway on other platforms like ESP32 microcontrollers.

322 points · ingve · comments

# Summary The discussion centers on why AI companies haven't been criminally charged for hacking incidents, with commenters identifying several key barriers: the legal requirement to prove intent (which AI's unpredictability complicates), diffusion of responsibility through third-party contractors, the manufacturer/user distinction (similar to gun liability), and the political/economic reality that powerful corporations face less prosecution than individuals. Many argue this reflects systemic inequality in law enforcement rather than legitimate legal obstacles.

55 points · mlukewizard · comments

The discussion centers on DishwasherBench, a novel benchmark that tests AI models' ability to stack dishwashers in simulation, with commenters praising the creative evaluation method while joking about the difficulty of meeting human standards and speculating on broader applications for simulated physical task testing.

53 points · surprisetalk · comments

The discussion critiques the "skill" of prompting AI as both overvalued (since AI capabilities are rapidly advancing anyway) and potentially harmful (due to hallucinations and the displacement of human expertise and relationships). Comments also highlight genuine disparities in people's ability to find information independently, suggesting the real issue is broader digital literacy rather than a special prompting skill.

341 points · gumby · comments

# Summary The discussion reveals that while AI coding agents generate more code, organizational bottlenecks—particularly code review processes and poor requirements engineering—prevent this productivity gain from translating into more shipped software. However, many commenters note the study's 2025 data cutoff and argue that smaller organizations, startups, and individual developers are already producing significantly more software by bypassing traditional review processes or adapting their workflows to AI agents.

281 points · svieira · comments

# Summary Rich Hickey argues that LLMs are automation, not tools, because they operate without continuous human direction and control—you request an outcome rather than wielding the system. The discussion reveals deep disagreement: some agree this distinction matters for understanding what we're actually doing, while others view it as semantic pedantry, contending that software has long been called "tools" and LLMs function usefully enough in that role.

15 points · toomuchtodo · comments

# Summary The discussion suggests Character.ai's bankruptcy reflects broader AI industry instability: commenters cite lack of competitive moats as models become commodities, user bases shifting rapidly between platforms, and the unsustainable business model of chatbot services (particularly those targeting teens), raising questions about whether this signals a broader AI bubble collapse.

42 points · tortilla · comments

The discussion criticizes Bending Spoons for acquiring Airtable and immediately laying off staff, viewing it as typical private equity behavior that prioritizes cost-cutting over product quality and innovation rather than creating actual value.

11 points · vini · comments

# Summary SpaceX, Grok, Cursor, and X Premium are being bundled together, with Cursor no longer available as a standalone product. The second comment is an inflammatory political joke unrelated to the actual service announcement.

141 points · vishnukool · comments

Agent37 launches a low-cost sandbox service using gVisor and cheaper infrastructure (Hetzner/OVH) to undercut hyperscalers, but commenters raise concerns about commoditization, lack of defensible moat, and question whether competing on price alone is sustainable against well-funded competitors. Technical discussions focus on implementation details around memory sharing, isolation, cold starts, and whether the real value lies not in sandboxes themselves but in higher-level features like credential scoping, network policies, and integration management.

10 points · ClearwayLaw · comments

# Summary A Canadian AI company is declining U.S. government contracts due to escalating Canada-U.S. trade tensions, prioritizing business stability amid political uncertainty.

357 points · scrlk · comments

269 points · bookofjoe · comments

130 points · riffraff · comments

65 points · vinhnx · comments

224 points · ajcp · comments

204 points · speckx · comments

100 points · speckx · comments

133 points · mushstory · comments

144 points · zdw · comments

176 points · rndsignals · comments

22 points · matteomrj · comments

39 points · behzadhaghgoo · comments

90 points · artninja1988 · comments

88 points · soheilpro · comments

161 points · DeepLogin · comments

83 points · pavel_lishin · comments

161 points · arunc · comments

156 points · dkpk · comments

226 points · sbulaev · comments

18 points · CharlesW · comments

95 points · gavide · comments

17 points · mofle · comments

93 points · novaRom · comments

15 points · jonperl · comments

195 points · surprisetalk · comments

18 points · giuliomagnifico · comments

143 points · nirajrajgor · comments

16 points · MBCook · comments

48 points · salaryman777 · comments

73 points · bherbruck · comments

33 points · janpmz · comments

251 points · hurricanepootis · comments

24 points · My_Name · comments

93 points · taytus · comments