Today’s thread is AI moving from abstraction into policy, budgets, and labor: Uber’s $1,500 monthly cap offers a concrete signal for enterprise AI pricing, Ted Chiang pushes back on claims that language models are conscious, and Meta’s limited pause on workplace tracking shows how AI is normalizing white-collar surveillance. Security and infrastructure follow, with a one-click GitHub token theft bug in web VS Code and Let’s Encrypt’s plan for post-quantum certificates. Elsewhere, Elixir adds gradual typing, AI demand keeps squeezing PC memory prices, mathematicians warn about AI’s effects on training, and DaVinci Resolve 21 expands its editing and photo tools.
-
Uber's $1,500/month AI limit is a useful signal for AI tool pricing
-
simonwillison.net
- 767 comments
Simon Willison argues that Uber capping AI coding spend at $1,500 per engineer per tool per month is a rare real-world pricing signal for enterprise AI. The comments treated it less as a pricing benchmark than as evidence that unlimited “token maxing” is ending and companies are starting to budget AI like any other expensive software input.
-
-
Artificial intelligence is not conscious – Ted Chiang
-
theatlantic.com
- 1368 comments
Ted Chiang’s Atlantic essay argues that today’s large language models are not conscious and that companies like Anthropic muddy the issue by talking about chatbot “well-being” as if fluent text generation implied inner experience. The comments mostly agreed the consciousness claims are overblown, but they split sharply on whether Chiang’s core argument actually rules out machine understanding or just restates old “stochastic parrot” objections in cleaner prose.
-
-
Meta workers can opt out of being tracked at work up to 30 min
-
bbc.com
- 743 comments
A BBC report says Meta staff can ask for 30-minute pauses from new workplace tracking systems that monitor computer use for AI training and security. The comments treated that "opt out" as cosmetic at best and a signal of how normal white-collar surveillance is becoming, especially when AI makes it cheap to analyze everything.
-
-
1-Click GitHub Token Stealing via a VSCode Bug
-
blog.ammaraskar.com
- 101 comments
A security researcher showed that a bug in the web versions of VS Code used by GitHub could let a malicious repository or redirected page trick a user into installing an extension and stealing their GitHub token with one click. The comments focused less on the exploit mechanics than on a broader failure: github.dev gets a broadly trusted GitHub session by default, which turns any extension or webview bug into an account-wide problem.
-
-
A Post-Quantum Future for Let's Encrypt
-
letsencrypt.org
- 166 comments
Let's Encrypt outlined how it plans to issue post-quantum web certificates using Merkle Tree Certificates, a redesign meant to keep TLS handshakes small even though post-quantum signatures are much larger than today's. The comments treated the direction as necessary, but zeroed in on the operational cost of landmark syncing, hybrid migration details, and whether signatures are actually as urgent as key exchange.
-
-
Elixir v1.20: Now a gradually typed language
-
elixir-lang.org
- 409 comments
Elixir 1.20 ships the first stage of a gradual type system that works mostly through existing pattern matching and guards, without adding new syntax or runtime overhead. The reaction was largely enthusiastic because it promises earlier bug detection while preserving Elixir’s long-standing strengths on the BEAM, especially stability, fault tolerance, and backwards compatibility.
-
-
32GB of DDR5 now costs $375 – AI shortage continues to squeeze PC building
-
tomshardware.com
- 391 comments
A Tom’s Hardware piece says desktop DDR5 has spiked to about $375 for a basic 32GB kit, with commenters reporting similar jumps across DDR4, SSDs, HDDs, and server memory. The core signal is that AI datacenter demand is now squeezing not just gamers but small businesses, refurb markets, and anyone buying memory-heavy systems.
-
-
I was recently diagnosed with anti-NMDA receptor encephalitis
-
burntsushi.net
- 248 comments
BurntSushi published a personal account of being diagnosed with anti-NMDA receptor encephalitis, a rare autoimmune brain inflammation that first looked like anxiety and psychosis and led to a psych ward before neurology caught it. Readers used it to talk about how often serious conditions get mislabeled as psychiatric, why timelines and patient advocates matter, and where AI can and cannot help with hard diagnoses.
-
-
Mathematicians issue warning as AI rapidly gains ground
-
science.org
- 349 comments
Science covered the Leiden Declaration, a statement from mathematicians warning that rapid AI progress in research math could hollow out training, flood peer review with junk, and shift the field toward machine-generated results that humans no longer understand. The comments mostly agreed the real risk is not wrong answers but losing the human pipeline that turns students into researchers, though many also said AI will still be a powerful teaching and discovery tool.
-
-
DaVinci Resolve 21
-
blackmagicdesign.com
- 274 comments
Blackmagic released DaVinci Resolve 21, a major update to its video editor that also adds a new photo workflow many readers see as a real Lightroom challenger, especially on Linux. The thread largely shrugs at the AI branding and focuses on the practical upside: a powerful free tier, a rare one-time paid license, and enough new photo and motion graphics features to pressure Adobe in several categories.
-
-
AI outperforms law professors in Stanford Law study
-
law.stanford.edu
- 357 comments
Stanford Law publicized a study claiming AI beat law professors, but the actual setup was narrower: law professors blind-rated AI and human answers to first-year contract-law tutoring questions and preferred the AI answers about 75% of the time. The comments focused less on whether LLMs are useful in law and more on whether this study mostly measured polished, confident writing under a favorable evaluation design.
-
-
Pwnd Blaster: Hacking your PC using your speaker without ever touching it
-
blog.nns.ee
- 121 comments
A security researcher showed that a Creative Sound Blaster Katana V2X soundbar can be reflashed over Bluetooth Low Energy without pairing or meaningful authentication, then turned into a malicious USB device that types commands on the connected PC. Readers were struck less by the hack itself than by Creative reportedly saying this "does not present a cybersecurity risk" and declining to treat it as a vulnerability.
-
-
ESP32-S31
-
espressif.com
- 200 comments
Espressif released the ESP32-S31, a new dual-core microcontroller with Wi-Fi, Bluetooth LE Audio, Ethernet support, and a shift to RISC-V in a higher-end ESP32 family part. Readers were excited less by the raw spec sheet than by what it means for tooling, Rust support, DSP-style workloads, and a likely new default chip for connected embedded products.
-
-
MacBook Neo is so popular that Apple doubled production
-
macrumors.com
- 505 comments
A MacRumors report said Apple has doubled MacBook Neo production after stronger-than-expected demand for its new $599 entry laptop. Commenters largely saw that as proof Apple finally found a low-end price point where build quality, battery life, and support burden beat Windows laptops and even used Macs for mainstream buyers.
-
-
Leiden Declaration on Artificial Intelligence and Mathematics
-
leidendeclaration.ai
- 80 comments
A group of mathematicians and related researchers published the Leiden Declaration, a statement on how artificial intelligence should be used in mathematics and how institutions should respond. The comments mostly treated it as a serious attempt to defend proof reliability, attribution, and training in a field where AI can generate convincing but hard-to-check work at scale, though some saw it as protection of old academic norms.
-
-
PlayStation Architecture
-
copetti.org
- 69 comments
A widely praised deep-dive on the original PlayStation’s hardware sparked a nostalgia-heavy discussion about how the console really worked, from weird memory mapping and branch-delay quirks to why PS1 graphics still look so distinctive. The useful signal was not just that the article is good, but that many of the console’s signature artifacts came from concrete architectural compromises that developers and emulator authors still have to account for.
-
-
Stop Killing Games
-
jxself.org
- 272 comments
A blog post argues that the "Stop Killing Games" campaign points to a deeper problem with proprietary software and says games should follow free-software freedoms so players can keep using, modifying, and sharing them after publishers walk away. The comments mostly rejected that jump from preservation to full software freedom, but surfaced a sharper split between practical end-of-life fixes for online games and a broader ideological push to make games open source.
-
-
Every Byte Matters
-
fzakaria.com
- 151 comments
A blog post uses a Java game-style example to show how memory layout changes performance, especially when scanning millions of objects. The useful takeaway from the comments is that this is really about choosing array-of-structs versus struct-of-arrays for a specific access pattern, not a universal rule that "every byte matters."
-
-
Gemma 4 12B: A unified, encoder-free multimodal model
-
blog.google
- 394 comments
Google released Gemma 4 12B, a 12 billion-parameter multimodal model that takes text, images, and audio while dropping the usual separate vision encoder in favor of a much simpler input projection. Commenters saw the main appeal as local use on midrange hardware, but the thread quickly converged on caveats around quantization, uneven vision quality, and fuzzy marketing about what actually fits in 16GB.
-
-
New Texas Instruments 5532 chips are not the 5532s we’ve used for decades
-
groupdiy.com
- 54 comments
A long-running audio op-amp part number, the NE5532, appears to have been materially changed by Texas Instruments without a new name, including lower maximum supply voltage and lower slew rate. The reaction is that this is not a minor process tweak but a compatibility trap for anyone treating a “5532” as a stable, interchangeable commodity part.
-
-
Ableton Extensions SDK
-
ableton.com
- 96 comments
Ableton released an official Extensions SDK for Live that lets developers build JavaScript and TypeScript tools with custom UI inside the DAW. Early users say it looks great for panels, utilities, and external-service integrations, but it is not a replacement for Max for Live or a real-time audio API yet.
-
-
Pluto.jl 1.0 release – reactive notebook for Julia
-
discourse.julialang.org
- 39 comments
Pluto.jl 1.0 is the first stable release of a Julia notebook environment built around reactive execution and reproducible, self-contained notebooks. Readers liked the model and the educational polish, but the comments also made clear that Pluto’s strongest ideas come with opinionated workflow choices that still limit broader day-to-day use.
-
-
Roku LT Operating System open source distribution
-
blog.roku.com
- 70 comments
Roku published an open source distribution of the small operating system used in some of its embedded devices, with code on GitHub and video-first tutorials. Readers liked seeing real C code and docs in the repo, but most of the attention went to the gap between "open source" and actual user control over Roku hardware and data collection.
-
-
The Unreasonable Redundancy of Nature's Protein Folds
-
research.ligo.bio
- 59 comments
A research blog post argues that natural proteins reuse a surprisingly small set of 3D shapes, or folds, across many different functions, instead of spreading across a vast possible design space. Readers found the result directionally familiar to biochemists, but useful as a sharper framing of a bigger question: whether protein fold space is intrinsically small or whether evolution has only explored an accessible corner of it.
-
-
What I've learned about the trombone
-
bryanhu.com
- 81 comments
A blog post from an adult amateur trombonist explains how the instrument works, from slide positions and harmonics to intonation and glissando, in plain language for non-musicians. Commenters pushed past the intro and got into the real substance: why trombones are hard to tune, how brass players actually adjust pitch in ensembles, and what parts of the post were oversimplified or technically off.
-
-
More than 6 out of 10 people turn to AI for psychological support
-
axa.com
- 91 comments
AXA published a global survey claiming more than 6 in 10 people have used AI for mental health questions or support, and many commenters immediately doubted the number and the framing. The useful signal was less "AI therapist" adoption than the narrower ways people already use chatbots as journaling tools, pep talks, and low-cost coaching.
-
-
Writing Portable ARM64 Assembly (2023)
-
ariadne.space
- 27 comments
A blog post explains how to write ARM64 assembly that works across Apple Silicon and other 64-bit ARM systems by sticking to the common ABI and syntax subset. Commenters mostly agreed the technique is useful for leaf routines and optimized kernels, but said the title overstates portability because Windows, OS-specific registers, and system interfaces still break the abstraction.
-
-
Rootshell: A new E2EE email service hosted in Iceland
-
rootshell.is
- 48 comments
Rootshell is a new email service that says it offers end-to-end encrypted email from servers in Iceland. The comments mostly challenged whether it is actually end-to-end encrypted, then piled on with sign-up failures, trust concerns, and several concrete mail security misconfigurations.
-