|
📖 Read In Depth
|
Hypothesis, Antithesis, synthesis
Antithesis — the autonomous testing/verification company — published a post connecting Hegelian dialectics to their approach to software correctness. For someone who builds algorithms from scratch to understand them, this intersection of philosophy of knowledge and software engineering craft is worth reading carefully.
hn/Best Stories
|
Show HN: Gemini can now natively embed video, so I built sub-second video search
A Show HN demonstrating Gemini Embedding 2's native video-to-vector projection — no transcription or frame captioning, just raw video embedded directly into a 768-dim space alongside text queries. A concrete proof-of-concept of a genuinely new multimodal capability worth understanding at the implementation level.
hn/Best Stories
|
Yann LeCun’s New LeWorldModel (LeWM) Research Targets JEPA Collapse in Pixel-Based Predictive World Modeling
Yann LeCun's new LeWorldModel research targets the collapse problem in JEPA-based pixel-level predictive world modeling. This is directly relevant to the ongoing LeCun vs. scaling-law camps debate — worth digging into for anyone tracking alternative architectures to transformers.
reddit/r/singularity
|
U.S. and Iran Are Talking About Ways to End Their War: What to Know
The most comprehensive overview of the US-Iran war situation: ongoing diplomacy, Trump's contradictory signals, Strait of Hormuz disruptions, and the oil price implications. Relevant backdrop for understanding the energy/macro environment affecting everything from compute costs to the broader economy you live in.
nyt/Top Stories
|
|
🎬 Check It Out
|
In Marathon, A.I. Is Everywhere and Humanity Is an Anomaly
A piece on 'Marathon,' a game set in a world where AI is ubiquitous and humanity is the anomaly — tracing how AI depiction in games has evolved from sci-fi villain to systemic infrastructure. Worth a look if you want fiction that actually engages with the texture of AI-saturated society rather than retreading Terminator tropes.
nyt/Arts
|
|
⚡ FYI
|
Goodbye to Sora
OpenAI is shutting down Sora, its AI video generator, just three months after signing a multiyear Disney deal. The HN thread is the most substantive source — commentary points to open-source video models outcompeting it and OpenAI redirecting compute toward agentic/coding products as the real drivers.
hn/Best Stories
|
How is Gemini 3.1 at the top of SWE-bench?
Thread questioning the gap between Gemini 3.1's SWE-bench rankings and real-world developer experience compared to Claude Opus 4.6 and GPT 5.4. Classic benchmark-vs-deployment divergence discussion — useful signal for someone choosing tools for serious coding work.
reddit/r/singularity
|
New comment by detente18 in "Tell HN: Litellm 1.82.7 and 1.82.8 on PyPI are compromised"
LiteLLM versions 1.82.7 and 1.82.8 on PyPI were compromised via a supply chain attack originating from their Trivy CI/CD integration. If you use LiteLLM in any ML infrastructure, check your dependency versions immediately.
hn/Best Comments
|
Anthropic in Contact With Professional Analytic Philosophers to Evaluate reasoning Capabilities of Models
Anthropic is engaging academic analytic philosophers to evaluate model reasoning capabilities — a meaningful signal about how they're approaching capability evaluation beyond standard benchmarks. Interesting from both an alignment and epistemology-of-AI-progress angle.
reddit/r/singularity
|
CEO of Figure.AI teases Hark, an advanced AI lab that aims to develop an AI capable of sensing and interacting like humans - "AGI, in the limit, should feel like a sci-fi movie"
Figure AI's CEO is launching Hark, a new lab focused on human-like AI sensing and interaction, claiming current LLMs 'feel incredibly dumb' for embodied tasks. Relevant to the hardware/embodied-AI track and how people who've shipped physical robots think about the gap between language models and AGI.
reddit/r/singularity
|
🚨 SpaceX aims to file IPO as soon as this week
SpaceX is reportedly filing its IPO prospectus within weeks, targeting a June listing that could raise over $75B with significant retail allocation. One of the largest potential IPOs in history — material for anyone thinking about compute infrastructure, aerospace, and the Musk empire's capital structure.
reddit/r/investing
|
What if AI demand shifts away from just GPUs
Discussion on whether AI inference demand is shifting from GPUs toward CPUs, citing ARM's new AGI-branded data center chip and claims of 2x performance per rack. Thin on detail but worth flagging given the GPU supply chain implications for ML infrastructure.
reddit/r/investing
|
Apollo Just Gave Investors Only 45% of Requested Withdrawals. BlackRock, Morgan Stanley, and Blue Owl Are Doing the Same Thing.
Apollo, BlackRock, and others are capping private credit fund withdrawals at 5% against 11%+ redemption requests — a liquidity stress signal in a sector that ballooned post-2020. Connects to the NYT Moody's downgrade piece and is worth knowing for anyone tracking macro credit conditions.
reddit/r/investing
|
Ripgrep is faster than grep, ag, git grep, ucg, pt, sift (2016)
The classic 2016 ripgrep benchmark post is trending again on HN. If you haven't read BurntSushi's deep dive into why rg beats grep (Unicode handling, SIMD, Rust's regex engine, smart ignore logic), it's a great example of performance engineering done right and explained with rigor.
hn/Best Stories
|