I bet this is on purpose. They want the models to mark enemies as killed when it didn't pull the trigger
I bet this is on purpose. They want the models to mark enemies as killed when it didn't pull the trigger
A digital CAPTCHA verification window titled "Select all squares with PIPES" against a plain white background. The window contains a 3Γ3 grid of numbered squares, mixing literal hardware, smoking pipes, and programming syntax.
These captchas just keep getting harder #rstats
LLMs β 3D Printers
this is wrong. llms don't output the average of [some category of text].
a prompt/context is a coordinate. stand in a crowded part of the map (generic prompt): get generic continuations. stand in a narrow pass w hard constraints (unusual & specific context) and u end up somewhere more interesting.
didnβt watch the video but this βweβre all one speciesβ line pisses me off bc itβs preemptive surrender
SO WHAT if we *were* different species! thatβs still not sufficient basis to create legal categories for different legal rights
if a being rocks up and says it can feel hope, itβs my peer
A few months ago software development changed forever. It's just taking people this long to properly try them out and see what they're capable of
100% of my code is written with #AI now.
I didn't plan for that. It just... happened. And honestly? There's grief in it. Years of building coding skills, and now I'm asking myself what they're worth.
But I've also never enjoyed building software more than I do right now.
New blog post!
Oh interesting, people who donβt know how to build software are getting mad at my post about building software. Cute.
Let me be clear, over the next year, the job of software engineer will shift dramatically to no longer have typing syntax into an editor as its primary time sink.
Great thread on why Linux age verification is being proposed by people who have never touched a Linux system
Do you have a chart that shows 5 years?
Global warming would not be happening if our grid operated off renewables and our transportation systems were electrified
Anthropic revenue growth is terrifyingly fast www.bloomberg.com/news/article...
It really is better, it's just that people hate adapting for some reason
Its also safer to have a car that quickly comes to a stop with no input
This is a great point and I'll add to it by saying a lot of people who claim they get carsick in EVs are really getting carsick because the driver of that EV is using one-pedal driving very badly.
It takes effort and fine pedal control to coast/drive smoothly which lots of drivers do not have.
Planning next moves
This just reeks of bad design. Why not wrap this in a tool that handles auth for the endpoints? Agents should never see the keys
βA two-panel internet meme comparing two types of machine learning updates using the standard "I want X," "We have X at home" format. βTop Panel: βHeader Text: Large, bold white text at the top reads: "Me: I Want Continual Learning." βVisualization: Below the header is a glowing illustration of a deep neural network. A steady, complex stream of colorful data points (icons representing lightbulbs, books, graphs, and gears) flows directly and seamlessly into the network's input nodes. βCaption: A caption below the visual states: "An AI system that learns incrementally and constantly adapts to new information." βBottom Panel: βHeader Text: Large, bold white text reads: "Mom: We Have Continual Learning At Home." βVisual Components: This section is split into two halves under the header. βLeft (Graph): A bar graph titled "DATA INPUT" showing clustered batches of data processing with a long, empty gap on the x-axis labeled "MONTHS OF NO UPDATES." βRight (Illustration): A tired-looking programmer with bags under his eyes is slumped in an office chair at a desk, looking at a computer screen. βScreen Details: The computer screen shows a classic "Batch Model Update Installer" window with a progress bar stuck at "5%" and text stating: "Processing batch from Q1 (Updating now: April)...". βAdditional Detail: On the wall, a calendar has many days crossed out with red 'X's, and the 10th day is circled in red with the text "MODEL UPDATE DAY!" scribbled next to it. βBottom Caption: Large white text at the very bottom concludes the meme: "The Continual Learning At Home:"
You can print the grids themselves. I don't have any models for them but I'm sure it's not hard to find
3D printing solves this
Bullshit Bench V2
new: 100 questions across several domains
- Anthropic & Qwen still on top
- Reasoning seems to hurt
- New models are *not* better than old (except Claude)
- Seems to be independent of domain
github.com/petergpt/bul...
We're happy to announce a long-term partnership with Motorola. We're collaborating on future devices meeting our privacy and security standards with official GrapheneOS support.
motorolanews.com/motorola-thr...
if people are using classified information to place bets, then theoretically others can analyze betting behavior to find signals that reveal classified information.
I love these weird tower robots
We just need to build more solar. All day every day, more solar.
No surprise this is in china
right, and stateful agents take this even further
coding agents actually are stateful agents. the code base is the memory. we just started doing it more directly and intentionally
Sam is a snake
almost every senior person you would want to hire in this industry is rich enough that they don't have to work for you, because they don't have to work for anyone. this leads to weird incentives in that they aren't fighting over static pools of labor but an elastic one where ethics are important.
How much OpenAI stock does the gov have?