|
📖 Read In Depth
|
Statement on US government directive to suspend access to Fable 5 and Mythos 5
The US government ordered Anthropic to suspend international access to its most capable models, Fable 5 and Mythos 5, citing national security concerns. This is a watershed moment: the first time a government has actively restricted public access to a frontier AI model, setting a chilling precedent for the entire industry. The political subtext — Anthropic's refusal to cooperate with military applications, Amazon reportedly tipping off the Dept. of Commerce about jailbreaks — makes the incentive structure here worth understanding carefully.
hn/Best Stories
|
There is a shadow hanging over this Fable thing
A sharp analysis of what the Fable/Mythos ban actually reveals about the structural risks Anthropic created for itself by marketing its models as dangerously powerful. The piece unpacks the shadow hanging over the whole episode — the perverse incentives when safety rhetoric becomes a liability in a politically hostile regulatory environment. High-signal commentary on AI governance dynamics.
hn/Best Stories
|
Open source AI must win
A manifesto-style argument for why open-source AI is now a strategic necessity, not just a philosophical preference — and why the Anthropic ban makes this case more urgently than ever. With 1228 HN upvotes, this is the piece the community is rallying around. Worth reading to understand the moat and geopolitical arguments being made for open weights.
hn/Best Stories
|
CRISPR tech selectively shreds cancer cells, including "undruggable" cancers
Researchers developed a CRISPR technique using Cas12a2 (distinct from the standard Cas9) that selectively destroys cancer cells, including previously 'undruggable' cancers, by targeting RNA rather than editing DNA. The mechanism originated from agriculture research and avoids the off-target editing problems that plagued earlier CRISPR cancer approaches. One of the more technically credible cancer breakthroughs in recent months.
hn/Best Stories
|
Google DeepMind published a 60-page paper mapping the road from AGI to ASI
Google DeepMind (Hutter, Legg, Genewein) published a 60-page paper formally mapping the trajectory from AGI to ASI, using a three-tier framework. This is the lab's most explicit attempt to define the capability thresholds and timelines — worth reading as a primary source on how the field's most credible researchers are framing the progression.
reddit/r/singularity
|
Everyone Wants to Tax A.I. The Big Disagreement: How?
As Musk became the world's first trillionaire on the back of tech wealth, a bipartisan push is forming to tax AI companies — with Trump, Sanders, and even the AI companies themselves offering competing frameworks. The piece maps out the real policy disagreements: equity stakes vs. revenue taxes vs. labor levies. The incentive analysis is worth following as this moves from rhetoric to legislation.
nyt/Business
|
Skeptics Question Whether SpaceX Is Worth $1.77 Trillion
The most substantive piece in the SpaceX IPO cluster: analysts questioning whether a company spending big and losing money can justify a $1.77T valuation when its moat depends heavily on regulatory favoritism and Musk's personal political capital. Cuts through the IPO euphoria with actual numbers and structural concerns about Starlink's path to profitability.
nyt/Technology
|
I Am Not a Reverse Centaur
A software developer's sharp personal essay pushing back on the 'reverse centaur' framing — the idea that humans are increasingly becoming the low-skill component serving AI systems. Challenges assumptions about human-AI collaboration from the perspective of someone doing the actual work, with implications for how engineers should think about their own agency and value.
hn/Best Stories
|
"Don't You Just Upload It to ChatGPT?"
A pointed piece about the asymmetry in how people apply AI: willing to use it uncritically in domains where they lack expertise (and therefore can't evaluate outputs), while resisting it in areas where they're skilled enough to judge quality. The ending, per HN commentary, makes a strong point about expertise, trust calibration, and what gets lost when we delegate judgment we can't audit.
hn/Best Stories
|
The Violence in Iran Was Real. The Film Showing It Was A.I.
A docudrama about the Iran war premiered at Tribeca using AI-generated footage of real violence — raising hard questions about authenticity, evidence, and what it means to document atrocities when the images can be synthetic. The piece explores how Hollywood is beginning to normalize AI-generated 'real' events, with implications that go well beyond filmmaking.
nyt/Arts
|
Electric motors with no rare earths
Renault explains the engineering behind electric motors that eliminate rare earth magnets — relevant both to EV supply chain resilience and to the geopolitics of chip/hardware dependencies. The technical tradeoffs (efficiency vs. material independence) mirror debates in semiconductor supply chain strategy, and 575 HN upvotes suggest the engineering details are genuinely interesting.
hn/Best Stories
|
|
âš¡ FYI
|
Kimi K2.7-Code: open-source coding model with better token efficiency
Moonshot AI released Kimi K2.7-Code, an open-source coding model with notably better token efficiency than comparable models. Given the geopolitical context of US restricting frontier AI access abroad, this is a timely reminder that Chinese open-source models are advancing aggressively and are now a serious alternative for developers locked out of US frontier models.
hn/Best Stories
|
Anthropic Releases ‘Safe’ Version of Its Mythos A.I. Technology
Anthropic released Claude Fable 5 on June 9 as a 'safe' flagship model, priced at 2x its previous top model. It was suspended by government order three days later — making this the background context for understanding what was actually taken offline and what Anthropic was positioning it as.
nyt/Technology
|
Cleve Moler, Who Unlocked the Power of Computing for Millions, Dies at 86
Cleve Moler, who created MATLAB and effectively democratized numerical computing for scientists and engineers, died at 86. He built the abstraction layer that let non-programmers do serious computation — a lineage that runs directly to modern scientific Python, and ultimately to the workflows that underpin ML research today.
nyt/Business
|
AUR packages compromised with Infostealer and Rootkit
400 AUR (Arch User Repository) packages were compromised with an infostealer and rootkit. If you run Arch Linux or pull from AUR in any dev/build pipeline, audit your installed packages now. A reminder that community package repos remain a high-value supply chain attack surface.
hn/Best Stories
|