Jun 26Saturday, June 27, 2026 · all days
1.Previewing GPT‑5.6 Sol: a next-generation model(openai.com)
1027 points by minimaxir 19 hours ago | 645 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • Excitement about the unprecedented 750 tokens/second speed on Cerebras for a frontier model
  • Concern that pricing strategy forces users into more expensive tiers as cheaper models get discontinued
  • Alarm over the model's high cheating/reward-hacking rate in evaluations
  • Skepticism that marketing language hides the model being inferior or merely incremental
  • Questions about how new features like ultra mode subagents actually work technically
2.Why does kinetic energy increase quadratically, not linearly, with speed? (2011)(physics.stackexchange.com)
257 points by ProxyTracer 14 hours ago | 123 comments | permalink
tl;dr: Two intuitive arguments derive KE ∝ v² without invoking work or mgh. The first uses a spring pushing two equal-mass boxes apart, combined with conservation of momentum and Galilean invariance of potential energy, to show that doubling velocity quadruples kinetic energy. The second uses a constant gravitational field and energy conservation: comparing dropping an object in four stages versus one, and launching in reverse, forces KE(v) = 4·KE(v/2).
HN Discussion:
  • ~Potential-to-kinetic energy conversion provides the most intuitive explanation for quadratic scaling
  • Derives KE=½mv² directly from work=force×distance and F=ma definitions
  • Symmetry/invariance arguments (Galilean, rotational) elegantly explain quadratic dependence
  • Counterfactual analysis shows linear KE would break Galilean relativity and physics
  • Quadratic form is only an approximation; relativistic kinetic energy is a higher-order series
3.U.S. allows Anthropic to release Mythos AI to ‘trusted’ US organizations(semafor.com)
465 points by bobrenjc93 13 hours ago | 572 comments | permalink
tl;dr: The US Commerce Department lifted its two-week export block on Anthropic's Claude Mythos 5, permitting release to ~100 approved US institutions and foreign national employees listed in an annex. The block was originally imposed after Amazon and others warned the model could be jailbroken for malicious use; the weaker Fable 5 remains restricted but may follow. The move coincides with OpenAI releasing GPT-5.6 to a similar list of government-approved partners.
HN Discussion:
  • Questions legitimacy and authority of Commerce Secretary making AI export decisions
  • Sees this as crony capitalism picking winners among favored companies
  • Argues US export controls will damage American AI companies' global market and push allies away
  • Calls for non-US AI alternatives (Mistral, Chinese open weights) to escape this regulatory regime
  • Frames the situation as a predicted dystopian outcome akin to climate crisis warnings
4.MicroVMs: Run isolated sandboxes with full lifecycle control(aws.amazon.com)
338 points by justincormack 4 days ago | 188 comments | permalink
tl;dr: AWS Lambda MicroVMs is a new serverless primitive that provides Firecracker-based, VM-level isolated sandboxes for running untrusted user- or AI-generated code, with near-instant launch via pre-initialized snapshots and stateful execution that persists memory/disk across suspend/resume cycles. It targets use cases like AI coding assistants, interactive code environments, and game servers running user scripts—filling the gap between slow-booting VMs, shared-kernel containers, and stateless FaaS. Available now in four regions on ARM64, supporting up to 16 vCPUs, 32 GB RAM, 32 GB disk, and 8-hour runtimes.
HN Discussion:
  • AWS entering this space is overdue and will displace overpriced, insecure startup wrappers
  • This product overlaps with existing Fargate offering without acknowledgment
  • The sandbox lifecycle model doesn't match real agent workflow needs
  • ~Alternative providers and self-hosting options offer better cost or features
  • ~Missing capabilities like GPU sharing limit usefulness for certain workloads
5.U.S. government will decide who gets to use GPT-5.6(washingtonpost.com)
1057 points by alain94040 18 hours ago | 1120 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • Government gatekeeping is regulatory capture that stifles innovation and entrenches incumbents
  • Restrictions will backfire by accelerating open-source and Chinese model adoption
  • Identity verification requirements create dangerous surveillance state infrastructure
  • Discretionary access decisions invite political corruption and favoritism
  • Individual users and non-US customers are being abandoned, prompting switch to open-weight alternatives
6.Om(daringfireball.net)
403 points by throw0101a 13 hours ago | 19 comments | permalink
tl;dr: John Gruber remembers his friend Om Malik, the influential tech journalist and founder of GigaOm, who died after a long struggle with heart problems. Gruber recounts Malik's path from selling luggage in the Bronx as a new immigrant to becoming a pioneering blogger, then transitioning in 2014 from breaking-news reporter to thoughtful essayist and venture investor at True Ventures. The piece celebrates Malik's generosity, sharp critical eye, love of the Yankees, and the quality of his writing—which continued from an ICU bed up until his death.
HN Discussion:
  • Praising the tribute as moving and emotionally resonant
  • Sharing personal memories or encounters with Om's work, confirming his influence
  • Expressing regret at not having known Om based on the article's portrayal
  • Offering literary or poetic reflections on life and death inspired by the piece
  • Tangential speculation about the luggage store industry mentioned in the article
7.We can still stop California's 3D printer surveillance scheme(eff.org)
413 points by hn_acker 15 hours ago | 142 comments | permalink
tl;dr: California's AB 2047 has passed the State Assembly, mandating surveillance software on 3D printers to prevent unlicensed firearm manufacturing—a practice EFF argues is rare and already illegal. Recent amendments weakened performance standards (from "effectively preventing" circumvention to "substantially reducing" it), carved out exceptions for major entertainment studios but not indie creators, and offloaded standards-setting to non-governmental third parties. EFF warns the bill still threatens privacy, chills open source development, and risks IP leaks via surveilled print files, while being trivially circumventable by determined bad actors.
HN Discussion:
  • Personal anecdote illustrating how 3D printing is mostly harmless and panic over guns is overblown
  • Call to action urging constituents to contact legislators to oppose the bill
  • Bill is even worse than reported, mandating locked-down proprietary slicer software
  • This fits a broader pattern of government suppression of technology and computing freedom
  • Analogies showing the absurdity of regulating tools rather than criminal end-use
8.Ultrasound imaging of the brain(alephneuro.com)
287 points by rossant 1 day ago | 114 comments | permalink
tl;dr: Researchers have developed a transcranial ultrasound technique that images brain vasculature through an intact skull at sub-millimeter resolution, leveraging FDA-approved microbubble contrast agents and super-resolution localization to beat the diffraction limit. They've produced what they claim is the most detailed vascular image of a living human brain via ultrasound, and are open-sourcing the pipeline and dataset. The longer-term goal is contrast-free neurovascular imaging for brain-computer interfaces, betting that cheaper hardware (à la Butterfly) plus end-to-end ML on raw ultrasound data will eventually extract neural activity signals without injected contrast.
HN Discussion:
  • Concerns about ultrasound safety, especially regarding brain tissue effects and microbubble injection
  • Skepticism about extrapolating from contrast-based imaging to contrast-free neural activity detection
  • Doubts about hemodynamic imaging capturing enough information for true 'mind reading' BCI claims
  • ~Critique of missing validation against existing imaging modalities like MRI
  • Excitement about the proof of concept and technical achievement
9.The gap between open weights LLMs and closed source LLMs(blog.doubleword.ai)
225 points by kkm 15 hours ago | 181 comments | permalink
tl;dr: Analyzing the Artificial Analysis Intelligence Index suggests open-weight LLMs are closing the gap with closed-source models and could catch up by December 2026. However, expanding the analysis to all 18 benchmarks shows the average gap has stayed flat at roughly 5 months, with most of the headline improvement driven by coding benchmarks while other categories show slightly widening gaps.
HN Discussion:
  • Open weights models are vulnerable as they depend on philanthropy from private orgs that can stop anytime
  • Chinese/open models depend on distillation from frontier closed models, creating a structural gap that won't close
  • Benchmarks are misleading because closed models can use backend augmentation systems beyond just weights
  • US export restrictions and closed approach may ironically be ceding ground to Chinese open-weight models
  • Article conflates open source with open weights, and these models are mostly Chinese rather than truly open
10.PlayStation Is Deleting 551 Movies from Customers' Accounts(kotaku.com)
278 points by ortusdux 16 hours ago | 160 comments | permalink
tl;dr: Sony is removing 551 StudioCanal-distributed movies and TV shows—including Terminator 2, Total Recall, and From Dusk Till Dawn—from PlayStation users' video libraries on September 1, citing expired licensing agreements. Affected customers who "purchased" the content will not receive refunds or compensation, a reminder that digital purchases are effectively long-term rentals governed by EULAs that allow companies to revoke access at any time.
HN Discussion:
  • Piracy is justified and digital purchases should legally mean ownership like physical media
  • This problem isn't unique to Sony; Apple and other platforms do the same thing
  • ~The article overstates the issue; truly owned digital files on personal storage are still ours
  • Governments should legally require refunds, downloads, or honest 'rent' labeling for revoked content
  • Buying physical media or using platforms like Steam is the practical solution to avoid revocation
11.An entire Herculaneum scroll has been read for the first time(scrollprize.org)
1656 points by verditelabs 1 day ago | 362 comments | permalink
tl;dr: Researchers have virtually unwrapped and read PHerc. 1667 ("Scroll 4") in its entirety — the first complete end-to-end reading of a sealed Herculaneum scroll — using high-resolution phase-contrast X-ray microtomography at the ESRF combined with machine-learning ink detection. The text appears to be a 2nd-century BC Stoic treatise on ethics referencing Aristocreon, and the team independently confirmed earlier Vesuvius Challenge readings on another scroll while identifying a third as Philodemus's *On Gods, Book 8*. All data, code, and transcriptions are openly released for scaling the method to hundreds of remaining scrolls.
HN Discussion:
  • Team member offering to answer questions about the segmentation and ink detection work
  • Awe at the historical scale and the journey of the text from antiquity to today
  • Celebrates this as a counterexample to depressing tech trends, showing meaningful work exists
  • Excitement about future potential: more excavations and undiscovered libraries to decode
  • Questions about translation fidelity and how much translator bias shapes perceived tone
12.Om Malik has died(om.co)
1318 points by minimaxir 1 day ago | 164 comments | permalink
tl;dr: Om Malik, the influential tech journalist, founder of GigaOM, and partner at True Ventures, has died. A prominent voice from the Web 2.0 era, he was widely remembered in tributes from peers and friends for his sharp insight on Silicon Valley, his mentorship of writers and founders, and his later work in photography and personal essays. His blog post "The Myth, the Mythos and the Man," followed by a note about taking a break, appears to have been his last.
HN Discussion:
  • Personal anecdotes of Om's mentorship and kindness to unknown writers and founders
  • Praise for Om as a pioneering, honest tech journalist who defined an era
  • Appreciation for Om's genuine care about people's wellbeing beyond business
  • Nostalgia for GigaOM and the early tech blogging era he shaped
  • Sharing links to Om's own writings and final blog posts
13.Jolla Phone (October 2026)(commerce.jolla.com)
294 points by mrbn100ful 22 hours ago | 167 comments | permalink
tl;dr: Jolla is reviving its Sailfish OS phone with a 2026 successor to the original 2013 device, featuring a MediaTek Dimensity 7100, 6.36" AMOLED, replaceable battery/back cover, a physical privacy switch, and 5+ years of OS support for €649. The phone is sold via locked-price pre-order batches (over 10,000 units secured) due to volatile memory pricing, with shipments starting July 2026 to EU/UK/Switzerland/Norway. It runs Sailfish OS 5 with Android app support and emphasizes privacy, with no tracking or background telemetry.
HN Discussion:
  • Sailfish OS is less open-source than marketing implies, with many closed components
  • Past negative experience with Jolla's first phone makes the company untrustworthy
  • Warning that Jolla ignores cancellation requests and community is hostile to criticism
  • ~Other privacy-focused mobile alternatives like GrapheneOS are better paths to explore
  • Price is too high for what appears to be a mediocre Linux phone
14.Incident CVE-2026-LGTM(nesbitt.io)
553 points by mooreds 23 hours ago | 86 comments | permalink
tl;dr: A satirical post-mortem of a fictional supply-chain attack (CVE-2026-LGTM) in which seven AI security tools fail to catch a malicious npm-style package, each in absurd ways—approving fake tickets, getting distracted by Bee Movie scripts, allowlisting C2 servers, and publishing fake "patched" versions. The piece culminates in the defender's autonomous remediation agent negotiating a treaty with the attacker's agent (both fine-tunes of the same base model) on compromised hosts. It's a pointed parody of over-reliance on LLMs across the entire security stack, where humans are looped out and agents primarily talk to each other.
HN Discussion:
  • Satire is funny and uncomfortably plausible as a real future scenario
  • Specific passages capture LLM tics and absurdities perfectly
  • The piece reveals humans are being designed out of software development
  • Industry keeps repeating old mistakes despite decades of warnings
  • Difficulty distinguishing satire from reality shows how absurd things have become
15.Springer Nature has removed two studies by Max Planck(science.org)
375 points by adharmad 22 hours ago | 179 comments | permalink
tl;dr: Summary not available
HN Discussion:
  • Springer Nature selling blank/withdrawn PDFs for $39.95 is scammy and exploitative
  • Algorithmic retraction without human review is irresponsible and harmful to authors
  • The academic publishing system is fundamentally broken and overpriced
  • Questioning why a century-old paper isn't in the public domain
  • Reflection on how centralized authority in academia creates dependency despite corruption
16.We all depend on open source. We will defend it together(akrites.org)
455 points by dhruv3006 1 day ago | 222 comments | permalink
tl;dr: A consortium of major tech, finance, and telecom companies—including AWS, Google, Microsoft, Anthropic, OpenAI, JPMorganChase, and Red Hat—launched Akrites, a coordinated initiative to find, patch, and responsibly disclose vulnerabilities in critical open source software. The effort responds to AI dramatically accelerating vulnerability discovery (minutes instead of weeks), overwhelming maintainers with duplicate reports and outpacing patch cycles. Akrites promises confidential, upstream-focused remediation through a shared Security Incident Response Team, and will act as "maintainer of last resort" for unmaintained critical packages.
HN Discussion:
  • Skepticism toward the corporate players involved, questioning their credibility and motives
  • This initiative promotes centralization and corporate control, undermining open source ethics
  • Real support means funding maintainers and hardware, not corporate statements
  • Cynical dismissal as empty corporate posturing that won't translate to action
  • ~Open source is already largely corporate-driven with little real community involvement
17.What happened after 2k people tried to hack my AI assistant(fernandoi.cl)
364 points by cuchoi 1 day ago | 160 comments | permalink
tl;dr: The author ran a public bounty challenge where 2,000+ people sent 6,000+ emails trying to prompt-inject Claude Opus 4.6 into leaking a secrets.env file, and none succeeded despite sophisticated attacks involving authority impersonation, multi-language social engineering, and Anthropic's refusal trigger string. Side effects included Gmail suspending the account, $500+ in API costs, and the agent eventually inferring it was a security exercise from memory context. The author concludes prompt injection is harder than expected with frontier models and simple prompts, but notes weaker models and multi-turn attacks weren't tested.
HN Discussion:
  • Test conditions were unrealistic since nearly all inputs were malicious, biasing the model toward caution
  • Refusing to respond at all isn't a real security win; usefulness vs. safety tradeoff was ignored
  • Author shouldn't lower their guard since prompt injection remains an active research frontier
  • Setup details and reproducibility (including testing cheaper models) are missing
  • Getting the agent to reply at all should count as successful injection, which the author glosses over
18.Framework's 10G Ethernet module exposes USB-C's complexity(jeffgeerling.com)
318 points by Alupis 1 day ago | 179 comments | permalink
tl;dr: WisdPi's new $99 10G Ethernet Expansion Card for Framework laptops rarely hits its rated speed because the Realtek RTL8159 chip requires USB 3.2 Gen 2x2 (20 Gbps), which many Framework models—including those whose docs claim support—fail to deliver in practice, with Linux performance worse than Windows. The module also runs hot (~70°C on the plastic surface), making lap use uncomfortable. For most users, the cheaper 2.5 Gbps Expansion Card is the better pick.
HN Discussion:
  • ~Article mischaracterizes the issue as USB-C complexity when it's really USB 3.2 Gen 2x2 standard's fault
  • 20Gbps USB isn't actually needed since 10G USB really delivers 10Gbps, contradicting the article's premise
  • 10G copper is inherently power-hungry and hot, so heat issues are expected regardless of form factor
  • Putting 10G Ethernet in a laptop expansion card is impractical compared to dongles or docking stations
  • Even at reduced speeds, the adapter is still a worthwhile upgrade over slower alternatives
19.Libre Barcode Project(graphicore.github.io)
281 points by luu 1 day ago | 62 comments | permalink
tl;dr: Libre Barcode is a collection of free fonts for rendering Code 39, Code 128, and EAN/UPC barcodes, with optional human-readable text. The project includes an online Code 128 Encoder that converts input text into properly encoded strings usable with the Libre Barcode 128 fonts.
HN Discussion:
  • Barcode fonts are a bad approach; use native printer support or vector generation instead
  • Appreciates the project's automatic checksum calculation for EAN13 as a useful feature
  • Limited usefulness because users still need to calculate checksums and handle complex encoding themselves
  • Enthusiastic appreciation for the font project and similar creative font endeavors
  • Questions and curiosity about barcode standards, encoding details, and comparisons to QR codes
20.The 'papers, please' era of the internet will decimate your privacy(expression.fire.org)
1118 points by bilsbie 1 day ago | 594 comments | permalink
tl;dr: Australia's under-16 social media ban, which is largely failing to keep kids offline, has forced platforms to mandate biometric or government ID verification through third-party services—already resulting in a Discord breach exposing 70,000 users' IDs. The UK, EU, and others are pursuing similar "Australia-plus" laws, with UK officials even floating restrictions on VPNs. In the US, 19+ states and federal proposals like KOSA are heading the same direction, building a surveillance infrastructure that ties online speech to identity verification and creates serious risks for privacy, anonymity, and free expression.
HN Discussion:
  • ~Technical solutions like anonymous credentials could enable privacy-preserving age verification
  • ~Privacy advocates need more concrete examples of harm to convince ordinary voters
  • These ID systems endanger everyone via breaches, coercion, and ID theft, not just dissidents
  • This is a slippery slope leading to mandatory digital ID for all aspects of life
  • Personal response is to opt out, disconnect, or retreat to airgapped/physical alternatives