hrpnk

clef from cloudflare runs on llama.cpp - being locked-in to strands cli would be a bummer and will slow down adoption.

Since it's a LoRa on Qwen, I assume this is runnable via llama.cpp. Pity that the PEFT/LoRa->GGUF translation is left to the user. Anyone got past:

    $ uv run --with transformers==5.19.0 convert_lora_to_gguf.py ~/Downloads/lora --dry-run --verbose
    [...]
      File "/Users/user/repos/llama.cpp/conversion/base.py", line 630, in map_tensor_name
    raise ValueError(f"Can not map tensor {name!r}")
    ValueError: Can not map tensor 'layers.0.linear_attn.in_proj_a.weight'

Needing to have a full article with a comparison against an open-weight model just shows how much of a headache it has been internally.

If you are forced to ID check with the bank via a 3rd party, the only way is to ask for removal of data after the ID check. Do you see other ways?

I wished OpenAI and Anthropic IPOd already. They just try to outcompete each other by finding the most splashy headlines undercutting prices in unhealthy and unsustainable means long-term.

Even if the trigger was spoofed, how come there is no secure channel that the govt provides to receive the data? Was this one also compromised?

What would the company need to do to comply with licensing? Release the studio, shared library, and server-side components as AGPL?

Even the simplest router that differentiates models between plan & execute will ensure that this is consistently followed. Folks have too high FOMO to choose themselves.

- write/update tickets & collect prompt queues for when it's back up

- read & respond to customer feedback

The breaking changes vs. Opus 4.8 are interesting [1]

1. Thinking on by default: On Claude Opus 4.8, requests without a thinking field run without thinking; on Claude Opus 5, the same requests run with adaptive thinking.

2. Disabling thinking is capped at high effort: You can still turn thinking off with thinking: {type: "disabled"}, but only at an effort level of high or below.

[1] https://platform.claude.com/docs/en/about-claude/models/migr...

The smell is about folks being attached to the same model because they're used to it. Just like any UI change in an app that one uses everyday is annoying.

What's often missing is the actual engineering. If you have evals for your use case, you use these to adjust to the new model. DSPy has great prompt engineering constructs it ships with.

The smell is all about vibing, where a model feels better because the structure of the answers is more familiar to a person, instead of engineering where given constraints and input/outputs one is using/bending the system to fulfil the requirements.

I love how there is a "Please do not discontinue gemini-2.0-flash[-lite], 2.5 is NOT an equivalent" from Feb 20th. Getting too attached to models is a smell.

They highlight the cache write price now much more in the guide. Did it increase vs. prior generations?

There is literally no more reason to use their service. I don't need a monthly expiring token pool drained at API pricing which I can get from any provider.

Wonder if this is a product decision or rather a limitation on GPUs. Their Inf2 instances require AWS Neuron [1]. Maybe porting models to run on their hardware takes too much time.

[1] https://aws.amazon.com/ai/machine-learning/neuron/

Rule files are in https://github.com/alibaba/open-code-review/tree/main/intern... (in Chinese)

If budgeted at $1,500/month per user, power users still can get 5-10x of that allocation if the user pool is large enough.

$1500/mth is token pricing.

Your other plans are fixed price with rate limits where you get more tokens than the dollar equivalent you pay monthly. These plans are economical only if majority of users spend less tokens in $ than the plan's costs. This subsidizes the gap vs. power users who spend multiple k$ monthly in API tokens.

Wonder if folks will flock to competitors that still over fixed-price review packages only for those companies to switch to usage-based pricing. They're next to be overrun with new customers and sit on the increasing bills.

What's the residual value of Copilot after the changes? For Enterprise plans, even copilot code reviews will be charged at token price + github action minutes for the execution of the review. One can roll one's own reviewer and have it spend the API tokens as well. At least one will be able to select the subset of files for the change set, if needed.

The background agents will also depreciate in value because of their harness that's a black box that's not optimized for token usage at all. Rolling one's own will be a better choice here.

Why the multipliers then at all?

"Paperasse is a collection of skills for AI agents (Claude Code, Claude Cowork, Codex, Mistral Vibe, Cursor, Windsurf, Cline, Aider) specializing in accounting, taxation, notary services and auditing of French companies."

I wish the team would be more responsive to popular issues - like inability to provide a dynamic api key helper like claude has. This one even has a PR open: https://github.com/anomalyco/opencode/issues/1302

Are there any videos or screenshots of readout? The webpage is very minimalistic.

https://github.com/dlupiak/claude-session-dashboard - I found this one today, it shows me nicely basic statistics. Do you know the similar tool that would be showing also the interactions of sub-agents? I want to see the user prompts and also statistics by file which tools have touched or read them.

The sample code is written in Elixir, however, they also release the full specs using which one can generate the application in the language of one's choice. I'm curious if this is really so straightforward.