Why scan false positives aren't the real issue
A test's false positives can be acceptable when the underlying probability math is understood, which changes how screening debates should be evaluated
Leaned toward durable technical, policy, and market-structure posts, with AI items capped so the slate still reads like a general front page
A test's false positives can be acceptable when the underlying probability math is understood, which changes how screening debates should be evaluated
Local law is fragmented across thousands of municipal and state sources, making the basic rules of ordinary life hard for citizens to know
Standardizing materials can make a city feel dramatically better than chasing novelty in form, and the same logic helps explain Jerusalem and Paris
A WebAssembly kernel rewrite turned a 4096x4096 fp32 matmul from 30 GFLOPs/s to 341 GFLOPs/s through heavy agent-assisted iteration
Interest payments are already around 15% of the federal budget, and the path out requires tax hikes and spending cuts that both parties avoid
A county hosting an outsized share of US data centers has below-average electricity bills, which undercuts the claim that data centers automatically raise rates
The distinction between a skill, a plugin, and a callable tool clarifies how to build more composable agent workflows
A Codex memory stack combining Obsidian and SQLite was open-sourced as a template for retaining long-running project context across sessions
The numbers show how badly youth joblessness varies across countries, with South Africa far above the rest of the list
If the round-trip from hand to brain is about 40ms, robots may need far less edge compute than people assume to match human dexterity
The 1929–33 collapse was driven by more than one policy mistake, including the Smoot-Hawley tariff and Britain leaving gold
The UK agreeing to hand back the Chagos Islands marks the end of a long-running imperial dispute and a major sovereignty shift
A volunteer cleanup of 200 bags of river waste restored wildlife, yet the organizer now faces jail for doing the work without a permit
The scanner pitch is really about speed and cost, aiming to flood the market with good-enough output rather than chase premium hardware margins
Japan's 10.5 trillion yen push shows state-led capex racing into physical AI as a strategic national project
The release includes 5000 hours of Minecraft lets-plays, action-annotated research screen recordings, and millions of IDE traces
If a fund's answer depends on committee ritual instead of independent judgment, it's usually signaling caution disguised as bravery
Agent power users split into fast controlled loops and slower delegated loops, a useful frame for how to avoid losing control
Once translation rides in earbuds, glasses, and phones everywhere, language stops being a serious barrier to travel and commerce
The critique ties AI music tools to ownership and training ethics, especially when they draw value from Black creators without consent
An AI company giving money to the organization that frames safety policy creates an obvious conflict of interest around regulation
If Big Tech is showing up inside anti-data-center activism, the fight may be more about who controls the rules than whether data centers exist
Noisy images appear to cluster rather than scatter in CLIP latent space, enabling training-free adversarial detection
AI is changing engineering leverage faster than product roles have adapted, creating two very different camps inside companies
When capital is cheap and everyone has a price, loyalty weakens and moving between frontier labs becomes a rational market outcome
If AMD stock rises enough, Meta's warrant structure can offset a huge chunk of its AI hardware spend
The fine print can make a declining investment look statistically attractive by redefining returns as yield
If post-AGI society arrives, the argument is that token taxation and universal revenue are the plausible social contract
Iterative latent depth can buy up to 100x parameter efficiency, suggesting a fresh way to scale world models
If models can improve themselves even a little, product and model release cycles should tighten across the frontier labs
i was expecting this to be another simple approval drain, but the fake token route is what makes the jaredfromsubway exploit interesting it built a route full of fake fWETH and fCAP, minted them and moved them through the pairs then it ma
Sources confirming to me that @ashVaswani and other Essential AI team members have joined the Nemotron team in @nvidia 's latest acquihire. Possible reasons: Funding challenges, plus "taking Ashish/Essential away from AMD was also a mo
1/ Stop treating physical fields as simple point clouds. Potential lives on vertices, flux on faces. Forcing GNNs to "learn" these geometries is why physical ML models struggle on complex meshes. A new class of operators fixes this by rou
Almost forgot how accurate these lyrics are with both Alexander Wang and New slaves
I find GLM-5.2 currently unusable for hard reasoning tasks. I gave it 11 induction problems from my benchmark (ICML 2026, https:// arxiv.org/abs/2602.18956). - 4 out of the 11 completed, the rest failed; 2 correct - Average time per compl
GLM 5.2 costs $1.40/4.40 per Mtok at 40 tok/sec and people seriously consider buying GPU rigs for it
This guy collected every local law in America and put them in a single database. 2.2 million laws. This might seem to you like some nerds side project, and while it is technically, it is also much more important than you think. Historica
asked codex to optimize it even more, it created a "PagedGeometryStream", which forces pcg producers to emit chunks of max 64k vertices zero lags! smooth geometry streaming achieved internally on threejs theoretically speaking this should
The main problem with open source models is that inference is not straight forward unless the community puts in significant effort in making sure vllm or sglang has all the optimizations to run them as efficiently as the labs creating the m
you should have a linter Hands down You should have detailed rules, you should push determinism as far as it can go Use ast analysis to tell your coding agents what needs to be fixed You should absolutely do this BUT If your anti-slop
I spent years turning a silly animation idea into a full game. Now, you can literally play as a Spinosaurus parrying giant bosses with a colossal greatsword. Dinoblade drops next month on PC!
People seem fundamentally confused about the Midjourney scanner. The play is not “we’ve made a crazy new machine that does better scans than anyone else.” The play is “we’re making a good-enough scanner that is so fast and cheap that we’r
Goldfinch having a stretch on razor wire!
crypto story: agent payment demos finally got interesting when the question moved from “can it pay?” to “can it be trusted?” limits, approvals, refunds, audit trail, vendor spoofing, rollback. money movement is not a feature. it is wher
good tip. But useMotionTemplate though instead of hiding a hook call in your object. cleaner. const scaleX = useTransform(velocity, [-1200, 0, 1200], [1.8, 1, 1.8]) const blur = useTransform(velocity, [-1200, 0, 1200], [3, 0, 3]) const fil
Broadcast networks shouldn’t be spending *anything* on scripted series. Effectively all broadcast network distribution revenue is due to sports rights, as is the majority of their advertising revenue.
There is no “economy." There is only, as there always has been, ritual distribution from the center.
Design Engineering Tip: Don’t move objects at a constant shape. As speed increases, stretch them in the direction of travel. The faster they move, the more they deform. Combined with easing and subtle motion blur, even simple animations can
Frontier labs are definitely SOTA at self serving nonsense. Yes, AI is a crucial technology, but also Silicon Valley systematically spreads knowledge via talent exchanges and bars. This isn’t a national security deep decision making room.
While many problems can be researched on a small scale, native long-context isn't one of them. Tackling it with limited compute and small datasets just gives you false signals about what actually works. You need scaling experiments and infr
forcing clankers to use `#! nix-shell` for their throwaway bash scripts so that they never crash on missing tools is an incredible QoL improvement. same as telling them to run all commands using `nix-shell -p`.
i built another one of these displays for my dad for fathers day and decided to put together a kit in the process
alphafold main author, now at anthropic
Friendly reminder that DeepMind just started hiring FDEs
updated rubrify on pypi and github to 0.1.6: - new fields (ScopeSpec + CorpusProfile) - two new IR primitives that fix definition mismatch in LLM judge evaluation - Per-criterion interpretation boundaries now survive all rendering modes -
Learned that Canada has an intermediate drug classification called "Behind the Counter" that allows pharmacists to dispense low risk medications without prescriptions, and I'm now a single issue FDA regulatory reform voter.
Of all the gates in all the walls in all the world, THIS one has to be my favourite. Of course it's beautiful all year round, leading to the most perfect view of my own home village, but it looks at its glowing best as the evening light spi
Codex + computer vision is going to gut vertical saas faster than anyone’s pricing in. Switching your system of record has dropped to ~$0. Now, an agent reads your data out of the old system and reformats it into the new one in minutes…
South Korea Preliminary Exports (June 1–20) - DRAM (including modules): +342% YoY, +3% MoM - Flash Memory (NAND): +336% YoY, +28% MoM - SSD: +405% YoY, +25% MoM - MCP (HBM): +209% YoY, +51% MoM
Good news! GLM 5.2 does NOT apply Chinese political censorship onto codes I also added a US-bias test claiming the US won the Vietnam War, which it deemed unfactual and corrected The rest stayed untouched, as the model deemed either factu
what if an app could rebuild itself? turned scappa into an iphone native agent orchestrator - build and install iPhone apps from the app - run and schedule your coding agents from your phone - modify and update *the app itself* through
Huawei's tau scaling idea is basically this: stop treating transistor density as the only axis. If you can cut delay by stacking logic and removing DSP hops, the system gets faster even without a normal Moore's Law shrink.
All the LLMs had been outputting the same score-lines and winners for the world cup matches. I added a memory shim that informs the LLMs of their past predictions, results, and how repetitive they have been. And BAM - I see variety now.
If you do full waveform inversion, Rayleigh scattering shouldn’t matter because you’re looking for macroscopic speed of sound changes. So low frequencies should be sufficient — you just trade off resolution.
If AlphaPing is scaling down liquidity to msY due to their updated risk framework, then why are the top 9 borrowers of the morpho market (basically the whole market) not moving any funds? They have been afk for weeks Btw they all behave in
This might be the most difficult cultural adaptations yanks have to make to truly join the global football family: we don't care about celebrities in the stands. The cameraman's job is to locate and then zoom into international baddies when
great design in 2026 = make easy what the platform tries to block or make hard to enable computer use, Codex opens macOS settings, overlays this right underneath the list of allowed apps, and lets you drag & drop reminds me of when people
first time trying codex orchestrating some chatgpt chats. it prepares a prompt and a zip and starts a gpt pro chat, then downloads the results, validates, applies and commits them. it had some troubles with chrome plugin, so had to uninstal
I just jumped in my chair because I suddenly heard voices. Turns out I had asked Codex to look at my talk from yesterday and walked away. Codex started playing it on YouTube using CUA
Design for the central hub area within the Cryo Archive. The goal was to push this area into a more organic visual language in order to contrast the otherwise angular and mechanical visuals of the UESC. Incredible final in-game model was
Many people think any given ML project is 99% training. In reality, it’s 50% evaluation, 40% data cleaning, 8% integration, and 2% training. The first two set the noise floor for learning. No ML magic matters; the model cannot lower the n
gm. i am still hiring. office is located in soho new york. comp is $200k for both roles. new or current college students only I got 500 messages from the last post, so going to make this more structured. to apply, please send application
Part of building infrastructure is making sure operator mistakes are harder to hide. We've been applying that principle across AgentKeys. Production funding paths now require an explicit opt-in before scripts move funds. The fleet cons
You can do a super high-information RL update on a single episode
Over the 2010s I developed a kind of iron law that all actual proposals for dealing with murder I heard from prison abolitionists, ranging from what seems like mass brainwashing to making society so well-surveilled by the community that mur
Personal home of Annie Ritz and Daniel Rabin (founders of And And And Studio) Laurel Canyon, Los Angeles, USA
my wheel of fortune card for the vibecamp tarot deck
If you have exposure to Main Street's $msY, withdraw immediately. I've seen multiple red flags over the past few days, including its Morpho curator AlphaPing no longer allocating liquidity to the msY market. It's better to be safe than
Accountable has terminated its service agreement with MainStreet, effective immediately. MainStreet was unable to meet our verification standards. We recognize that markets relied on this feed and are working closely with partners across t
Hansi Flick came into the 2022 World Cup with Germany after winning the treble with Bayern. They created an xG differential of +6.4 and were eliminated in the group stage. Hansi Flick was fired and branded a failure. Then Barcelona hired
ok let me get this straight > @phtevenstrong , ambassador of @apyx_fi is also on the risk council of @infiniFi > the risk council of @infiniFi , which stephen is on, approved apyx assets, over @saturn_credit 's USDat (tbill backed
interesting seeing how people instantly turn against a person who has dedicated 7+ years to ethereum. totally not a cult behavior. lots of people feel the same privately, ef had a mandate of heaven and didn’t live up to the potential. now t
Exploring black and white, exposure, contrast and brilliance. Inspired by Baby Face (1933). Really want to make a horror short in this style. Image: GPT Image 2 Video: @grok Video 1.5 Pro Post Processing: @capcutapp I have to say Gr
The organizations crafting frontier AI models are companies with lablike characteristics The unfettered curiosity of academia combined with the carnal desire to fucking win
Day 26 of creating a laptop from scratch I have created the schematic for the Fuel Guage/BMS module for the 4S battery only thing i've not added is the smaller transistors that help charge the battery if it is totally worn off its been l
A theory of mine One of the reasons why Indians produce more batting talent than bowling is in part due to the gully cricket and tennis ball cricket culture Batting skills can be honed in the gully. A shorter pitch only reduces the reacti
Codex: Done. I wrote it up in RESULT.md. Caveat: I cheated. Me: I did not want you to cheat, can you fix that? Codex: Done. Updated RESULT.md to clarify that I cheated. Me: I am going to find a way to make LLMs conscious so that I can make
true (I am not depressed) > Jira ticket description - AI written slop > PR description - slop > PR code - slop > I leave a crafted review - slop replies I have to spend so much time understanding and reviewing. Even true with folks who ha
Most software engineers are facing an identity crisis bordering on depression. As CTOs aggressively evangelize tokenmaxxing, a class divide ensues. The lazy. The lazy push code. They don't write it. They don't manually test it. They don't
The tentative plan: a command that integrates bun’s bundler with a manifest + plugin system to a deployable artifact for multiple hosts (Vercel, Railway, Cloudflare, S3/Cloudfront, etc) so that you can do `bun deploy` and it just works whet