throwaway2027

"The model makes an honest mistake and mistakenly deletes $HOME instead."

Obligatory https://www.destroyallsoftware.com/talks/the-birth-and-death...

That's quite slow I'm getting 8-12 t/s on a 13 year old CPU. (Speed varies by context size and other settings who knows)

https://news.ycombinator.com/item?id=48354801

Maybe, let me ask my coding agent what he thinks about this.

You should assume by default for any AI agent that it will read anything. Even if you manually allow/deny and "restrict" it to a subdirectory I would still hold that assumption. Claude reads your ~/.bash_history too so when you ran something it can use that same command.

Statement of AI use. The proof in this note is entirely due to GPT 5.6 Sol Ultra and the writeup with Codex (with GPT 5.6 Sol).

Quick! Someone (a human) copyright and patent it. /s

Quite interesting to see them and Meta and others release before OpenAI supposedly is to release GPT 5.6 today, would it be better to release it before or after? Calm before the storm type of thing?

While I believe they would delete a user's games, I don't know if they would actually willingly give up holding on to customer data.

So basically you get worst of both worlds, great.

You would obviously be trading storage for compute and time to retrieve the storage.

I'm not saying you're wrong but based on? Vibes?

I loved Nano Banana Pro. Are there local alternatives yet? I heard about Qwen Image, Klein and recently Krea, any suggestions?

Why would it matter? It was of benefit to someone I assume and even if the code is AI generated I think in most cases the results were positive so in theory next AI training runs should be able to learn from those generated results altough that might require work in how training is done or more aggressive filtering. It's literally free RLHF training.

Claude Code does it the same way (which doesn't excuse it obviously) but still.

curl -fsSL https://claude.ai/install.sh | bash

https://code.claude.com/docs/en/quickstart

I'm perfectly capable of programming without internet or AI but I would admit it would take longer and in the modern world we live in it's often not economical to do so. After programming for over 20 years you start to get in that flow automatically at least you used to do so. I don't know if people starting out to program will be able to, but most experienced developers will feel this way I assume.

Is it that time again?

Well at least you didn't submit an ID and selfie yet so you could make a new account right, the future is bright.

E-mail from Anthropic Team:

Hello,

We're writing to inform you about some updates to our Privacy Policy.

These changes only affect consumer accounts (Claude Free, Pro, and Max plans). If you use Claude Team, Claude Enterprise, the Claude Platform, or other services under our Commercial Terms or other agreements, then these changes don't apply to you. What's changing?

Claude can do more than ever — taking on bigger tasks and connecting with the apps you use. We've updated our Privacy Policy to be clearer about the data we collect and how we use it. We encourage you to read the updated Privacy Policy in full, but we’ve set out a summary of the key changes below:

1. Multi-step tasks and connected apps. As Claude takes on more multi-step tasks and works with third-party apps and services, we've explained the data this involves — including how data can flow to and from third parties when you connect a service or have Claude do tasks on your behalf.

2. Verification data. As part of our measures to keep our services safe and secure we may ask you to verify your age or identity, and we've described what we collect and how.

3. Study participation. If you take part in Anthropic studies, surveys, or interviews, we've explained the information we collect.

4. Additional information about our data practices. We’ve provided more detail about how we communicate with you and promote our services, including providing tailored recommendations about our services that may be of interest to you. We've also clarified the circumstances under which we may receive or provide data to third parties, and the legal bases we rely on when processing your data.

While our products have evolved, our commitments haven't: We don’t sell your data, Claude remains ad-free, and you can control whether your chats and coding sessions are used to train and improve Anthropic’s AI models. Learn more

For detailed information about these changes:

    Review the updated Privacy Policy
    Visit our Privacy Center for more information about our practices

Will try it when my limit resets.

I received this e-mail today and coincidentally some of my posts finally got some engagement and I guess you just get flagged as a bot by the bot system, welcome to the new internet. So much for open algorithm, it doesn't help you here.

X

Hello {username},

Your account, {username}, was reported and has been suspended for violating the X Rules.

Specifically, for:

Violating our rules against inauthentic behaviors.

You may not use our services to engage in inauthentic activity that undermines the integrity of X.

Note that if you attempt to evade a suspension by creating new accounts, we will suspend your new accounts. If you wish to appeal this suspension, please contact our support team.

You can also pursue alternative forms of redress, including out-of-court dispute settlement or judicial redress. Learn more here.

If you have an active X Premium subscription, it will not be automatically canceled by X. To cancel your X Premium subscription, follow these instructions.

Glad to see other people realizing this. I've been running Gemma 26B-A4B Q4 on a 2012 Xeon with 16GB to 24GB of RAM in a container. It's getting around 8 to 12 tokens per second. Obviously it's not comparable to huge contexts and running it on a GPU and the image decoder in llama.cpp is super slow compared to a GPU but for some small automation tasks and general trivia questions it's decent. The speed is just enough to not have to wait for it to finish so you can read along.

Here's my setup. You may want to figure out what the best optimizations are for your specific CPU like AVX2 because mine didn't have most of them. I did try MTP briefly but I wasn't getting performance improvements. You could play around with the batch sizes for cache or context or go even lower for Q2 and don't overcommit on threads either, but I would suggest either defaults or trying out llama-bench. This isn't by any means the best I assume but it worked decently for me and I sometimes swap out Gemma for Qwen. You could also lower q8_0 to q4_0 for more context but it could hurt quality some say, altough I have noticed it too on some models.

# Building

cmake -B build -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=ON -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DGGML_OPENMP=ON

# Running

export OPENBLAS_NUM_THREADS=4

export OMP_NUM_THREADS=4

OPENBLAS_NUM_THREADS=4 OMP_NUM_THREADS=4 \

llama.cpp/build/bin/llama-server -hf unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q4_K_XL --temp 1.0 --top-p 0.95 --top-k 64 --min-p 0.00 --jinja --host 0.0.0.0 --port 8080 --cache-type-k q8_0 --cache-type-v q8_0 --threads 4 --threads-batch 4 --ctx-size 8192 -n 8192 --batch-size 2048 --ubatch-size 512 --no-mmap --mlock --chat-template-kwargs '{"enable_thinking":false}' --no-mmproj -np 1 -fa 1

AI for me but not for thee.

"Cheap clients pay the least and complain the most."

I saw some small business owner complain about this behavior on twitter some time ago and he mentioned he only saw non-Americans do this and it made him really mad or something and he didn't provide the service and banned them or something. Funnily enough I do think this happens so sometimes I cancel instantly and sometimes deliberately wait until there are a few days left on the subscription exactly out of paranoia behavior that you'll get a worse service or something, that they must have some database field early cancel and mess with you or something.

Not to dismiss the AI but the important part is that you still need someone able to recognize these solutions in the first place. A lot of things were just hidden in plain sight before AI but no one noticed or didn't have the framework either in maths or any other field they're specialized in to recognize those feats.

Sad. I liked Gemini CLI. I used it a lot and occasionally use it these days. I've never tried Antigravity though.

December 2025 was the breakthrough for me. January Claude was euphoric, ChatGPT was up there. February Gemini cooked for a second there. March amazing. April the big bad nerf. May GPT 5.5 is just pure bliss altough 2x limits temporarily, not sure about Claude it's sort of okay still not as good as it felt before, slowly increasing limits with more compute and rebuilding good will.

That's probably why. You're not allowed to use the subscription with any other coding harness other than Claude Code.

The real danger is AI slop displacing actual good resources for software engineering. Once those are gone they will have to be reverse engineered from the AI it's weights.

I'm thoroughly enjoying using AI to write code, but it paid off by years of doing things the hard way before. I already was a so called "10x developer" if I speak for myself. I'm doing things even faster now with AI.

I already followed those rules mostly with StackOverflow and before AI.