Does this actually matter for multi-agent use cases? Surely people that are using swarms of AI agents to write code are just letting them resolve merge conflicts.
If you appreciate open source maintainers, detect when users are opening pull requests without human review and stop them. Feel free to keep burning their tokens, just stop making pull requests.
I would be concerned that laser cutting steel degrades the magnetic permeability around the cut. I plan to use a clamping jig designed for PCBs. I could always upgrade to a vacuum table.
This reminds me of Mrs. Davis (2023). An AI upends the world economy by giving people quests that are rewarded by fulfilling everyone’s inner-most desires. Each quest contributes to someone else’s reward. A nun sets out on a quest to destroy it. It is fully unhinged.
Prusa XL with higher temp filament, not enclosed. I was making parts that spanned corner to corner. It works fine once I prevent it from making 400mm linear runs.
3D printer beds have been getting bigger, but slicers don’t seem to account for curling as large prints cool. The problem is long linear runs on bottom infill and perimeters shrinking. I’ve been cutting my large parts into puzzle like shapes, but printing them fully assembled. This adds curved perimeters throughout the bottom layer, reducing the distance stress can travel before finding a seam to deform.
That said, a retro laptop this thick would look really nice in stained wood.
There are no office tools. It’s just a chatbot app. The page says they combined word and excel and PowerPoint, but it still doesn’t do anything but chat. I asked it to create a word document and it offered me a download link to a word template…
I checked way back machine and they have been making large changes to that page every day for the last month. It used to be a lie that office 365 was rename to office 365 copilot, yet it is an app with only chat bot functionality. They advertise the copilot integration for the main office apps now, but those are not part of the copilot app they are trying to trick you into installing.
Edit: The attacker didn’t actually know my email address. The google domain had a catch all email forwarding config that squarespace botched. Apparently they can spoof emails from my domain on a mailgun server as long as it’s to an email address on my domain.
MS actually changed their office.com landing page to a funnel that tricks you to into installing a copilot app. It used to be the dashboard for MS web apps. There are no links to the web apps, but they are all still there, you just have to know the subdomains. The app doesn’t have any of the functionality that page used to offer…
I have always been an advocate of forking, despite the overhead of maintaining patches, but porting patches should be trivial to automate now. There needs to be an easy way to publish, discover, and require community patches even if they don’t have the maintainer’s blessing.
I don’t understand why Thompson sampling would be used for an experiment that can be rapidly repeated. Isn’t reconfiguring the FPGA orders of magnitude slower than measuring latency? It seems like you could produce a high confidence posterior by resampling so fast that there is no point in prioritizing what needs to be revisited (especially with good priors).
Google just let through an email spoofed from my own domain (via a mailgun server). It was a phishing attack about the domain being shut down. The connection between the domain name and my personal email address have never been published. Either google or Squarespace leaked the info.
I have been working on a tool that aids in circuit tuning: model circuit equations as python functions, the solution space is discrete component values, auto solve for a target spec, build the circuits, record measurements, fit error, repeat until the experiment matches predictions. It adjusts nearly every parameter between tests and converges surprisingly fast. (25% to 2% error in 3 tests for an active band pass filter)
The MVP was hand coded, leaned heavily on sympy, linear fits, and worked for simple circuits. The current PoC only falls back to sympy to invert equations, switches to GPR when convergence stalls, and use a robust differential evolution from scipy for combinatorial search. The MVP works, but now I have a mountain of slop to cleanup and some statistics homework to understand the limitations of these algorithms. It’s nice to validate ideas so quickly though.
Dreams on PS4 had an SDF modeler, but I’m not sure if the runtime was SDF. Now that I think about it, the rendering engine had a Gaussian splat look to it years before that paper.
It looks like he fell backwards, implying the car hit him, but the bodycam actually detached. Until another bodycam shows why it fell, the car either clipped his vest or he pulled it off while drawing his gun…
Magpie-TTS needs a kernel compiled targeting Ampere, but it appears to be closed source. It was compiled for the 2018 T4, but not 2020-2024 consumer cards, just 2025 consumer cards.
Coffee stains should look like water color paints. The fluid deposits pigment more at dry boundaries as evaporation and absorption approach equilibrium.
I would test the CPU cooler since the fans ran so hard. Temps ramp up around the login screen, then stay hot and reboots get unpredictable.
I recently had a water cooler pump die during a Windows update. The pump was going out, but the unthrottled update getting stuck on a monster CPU finished it off.
I need to test the new audio demuxing model out for fan edits. Separating music, dialog, and sound effects into stems would make continuity much easier. Minor rewrites would be interesting, but considering Tron Ares botched AI rewrite dubbing so bad I’m not holding my breath.
I was recently trying to render a math visualization for a very large search space and fell into the map tiling rabbit hole. The idea was to randomly sample the viewport then pan/zoom to guide the search. OpenSeadragon was the only solution I got working, but it wanted the zoom pyramid prerendered and aggressively cached, so nothing updated. I don’t need the vector editor, so I’m tempted to port lightning to OpenSeadragon.
Sympy can (often) solve under constrained systems in terms of the free variables. The problem I run into is discrete constraints that make solving less closed form and more combinatorial search. When textbook amplifier formulas significantly diverge from physical reality I model the errors as linear correction factors and use gradient descent to correct it in a few experiments, but I’m curious if there is software that has solved this problem.
Plotly just shut off their Chart Studio web app and “replaced” it with a desktop app called Studio. That desktop app requires LLM chat input for every action. It uses this pattern and trying to figure out the magic words to make it do the basic tasks you have been doing with 3 clicks for the last decade is infuriating. Especially when you realize your data was local to your browser tab unless you saved it before and now it is unconditionally uploaded to a remote server with no obvious way to delete the data.