HN user
Incentive is for users in general to release sessions (sans PII, credentials) so all AI get better and there is alternatives. Even if China didn't do this, I don't see frontier labs being able to charge premium over others for long. RSI maybe?
OpenAI did this first.
In addition to safety training, automated classifier-based monitors detect signals of suspicious cyber activity and route high-risk traffic to a less cyber-capable model (GPT-5.2).
He seems to be saying that Claude Code can make discoveries. Does anyone think novel discoveries can be made from systems created by supervised learning only, and attempting to do so?
Claude-Code, which have brought true advances in ... programming. ... these systems have found things that are both novel and good.
Yet the Eternal September is what made the modern AI possible. I asked Cowork for share of the corpus before / after the event, and the corpus before it is <1%, which fits my hunch.
Many things, such as ksmbd seems ill-advised when looked at from security. New AI driven exploits era will likely make projects more wary to adding functions.
Laziness is a virtue, but when I asked GPT-5.4 to test scenarios A and B with screenshots, it re-used screenshots from A for B, defeating the purpose.
Are you mostly using the Composer model?
Copilot was there in AI based development first with tab completions.
Now, it may be the right call to immediately give up and shutdown after Opus 4.5, but models and subscriptions are in flux right now, so the right call is not at all obvious to me.
The agentic AI models could be commoditized, some model may excel in one area of SWE, while others are good for another area, local models may be at least good enough for 80%, and cloud usage could fall to 20%, etc. etc.
Staying in the market and providing multi-model and harness options (Claude and Codex usable in Copilot) is good for the market, even if you don't use it.
AI should decide the level of model needed, and fallback if it fails. It mostly is a UX problem. Why do I need to specify the level of model beforehand? Many problems don't allow decision pre-implementation.
Thx. In other comment, but setting Java options for proxy settings worked. I was guessing annotations would be based on UI framework borders(which I can toggle on screen via debug menu), but looks it's based on image recognition locally. Maybe you can have UI component borders as option(haven't heavily used it yet, so throwing around ideas)
MS showing "view summary" button for all meetings, then doing bait-and-switch to tell you to buy Copilot license (on a corporate seat no less, where regular users don't have purchasing decision power) is top annoyance now
Tryed both. Copilot said it was wrongly utf16, but it was utf8 according to vscode.
I got a workaround a la GH Copilot:
<pre>
android skills list
Picked up JAVA_TOOL_OPTIONS: -Djava.net.useSystemProxies=true
</pre>
Taking screenshots, optionally with component borders highlighted, and operating the UI with element names like "button1" instead of tap 200,30 looks useful. If I could get it to work.
The install command shown for Windows is 404.
`curl -fsSL https://dl.google.com/android/cli/latest/windows_x86_64/inst... | bash`
The URL shown for individual OSs work, but the script errors for me.
`curl.exe -fsSL https://dl.google.com/android/cli/latest/windows_x86_64/inst... -o "%TEMP%\i.cmd" && "%TEMP%\i.cmd"`
I manually downloaded the exe, but it say socket error. vibe coding is going strong!
Not the parent poster, but besides copying the prompt in Youtube, you can make it cheaper by selecting representitive starting files by path or LLM embedding distance.
Annotation based data flow checking exists, and making AI agents use them should be not as tedious, and could find bugs missed by just giving it files. The result from data flow checks can be fed to AI agents to verify.
gemini @YouTube is decent too.
Yeah, I thought Anthropic was for AI safety. Telling Ai to not be honest is a bad sign.
But once you go beyond that to less defined things such as code quality
I think they have a good optimization target with SWE-Bench-CI.
You are tested for continuous changes to a repository, spanning multiple years in the original repository. Cumulative edits needs to be kept maintainable and composable.
If there are something missing with the definition of "can be maintained for multiple years incorporating bugfixes and feature additions" for code quality, then more work is needed, but I think it's a good starting point.
I'm hoping Facebook will bring back API to access Groups. My family Photo is in it. I feeling trepidation because they failed to acquhire OpenClaw's author.
Maybe. People have run wildly insecure phpBB and Wordpress plugins, so maybe its the same cycle again.
The vocabulary has been long poisoned, but original definition of CSAM had the neccessary condition of actual children being harmed in its production. Although I agree that is not worse than murder, and this Claude's constitution is using it to mean explicit material in general.
I wonder if later challenges would be cheaper if summary of lesser challenges and solutions were also provided? Building up difficulty.
OpenCode has 11 options for installation.
I knew curl, npm, and docker, but asked Gemini about the rest:
bun, pnpm, yarn, brew, scoop, chocolatey, paru, mise.
Although its wonderful that people are building and creating, I also hope it calms down somewhat so I can choose from well tested few options in the future.
One surprising thing that codex helped with is procrastination.
The Roomba effect is real. The AI models do all the heavy implementation work, and when it asks me to setup an execute tests, I feel obliged to get to it ASAP.
Would some sparks fly when easy decompile of MSOffice and Photoshop are available, I wonder.
This is where I would guess the world destroing AGI/ASI will come about. The neverending cat-and-mouse game of ads/blockers driven by profit motive. LLMs will used by both sides in a escalating game, with humans with its attention and wallet stuck in the middle.
delivering 97% of the performance at 10% of the cost is a distraction.
Not if you are running RL on that model, and need to do many roll-outs.
The ideal case would be something that can be run locally, or at least on a modest/inexpensive cluster.
It's obviously valuable, so it should be coming. I expect 2 trends:
- Local GPU/NPU will have a for-LLM version that has 50-100GB VRAM and runs MXFP4 etc.
- Distillation will come for reasoning coding agents, probably one for each tech stack (LAMP, Android app, AWS, etc.)x business domain (gaming, social, finance, etc.)
Not complaining too loudly because improvement is magical, but trying to stay on top of model cards and knowing which one to use for specific cases is bit tedious.
I think the end game is decent local model that does 80% of the work, and that also knows when to call the cloud, and which models to call.