benob

5 ms·

1.
by benob 7d ago
Lots of people in non-tech jobs are dropping their laptop/desktop for their phone for all work-related tasks. They love having more screen real estate, plus their company pays for the premium.
2.
by benob 1mo ago
A natural next step is to use the reasoning traces to jailbreak the stronger models ( https://arxiv.org/pdf/2603.12277 )
3.
by benob 1mo ago
Maybe the correct UX could be to list the locations of settings data, their size and ask the user whether they want to leave them, put them aside in a dedicated folder ("the attic", "the basement" or whatever), or remove
4.
by benob 2mo ago
What matters for this injection strategy to work is to follow quite closely the style of the reasoning. It's particularly effective if you copy reasoning from the same context. If you cannot see the reasoning, you cannot duplicate it&#
5.
by benob 2mo ago
People will keep hiding reasoning because it allows prompt injection https://arxiv.org/pdf/2603.12277 , in addition to facilitating distillation (you don't pay the full cost of RL)
6.
by benob 2mo ago
The next step is to build software without bugs
7.
by benob 3mo ago
Longest definition and semi-columns are strong biases for right answer. Also, my run contained a lot of adjectives for which it is pretty obvious that noun definitions do not match.
8.
by benob 3mo ago
It may be a clever move. By using the same models as android (contractually?), they can compete on the user experience which they typically handle better than android phone providers.
9.
by benob 3mo ago
And papers on bias amplification in ML predate LLMs. I remember this specific one which was a spotlight paper at EMNLP: Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints, Zhao et al. https:/&#x2
10.
by benob 4mo ago
Does changing the date fix it?
11.
by benob 4mo ago
Deployed it to a huggingface space: https://huggingface.co/spaces/benoitfavre/needle-playground You can check the very simple docker file there.
12.
by benob 5mo ago
Here is llama-bench on the same M4: | model | size | params | backend | threads | test | t/s | | ------------------------ | ---------: | ---------: | ---------- | ------
13.
by benob 5mo ago
I get ~5 tokens/s on an M4 with 32G of RAM, using: llama-server \ -hf unsloth/Qwen3.6-27B-GGUF:Q4_K_M \ --no-mmproj \ --fit on \ -np 1 \ -c 65536 \ --cache-ram 4096 -ctxcp 2 \ --jinja \ --temp 0.6 \
14.
by benob 5mo ago
I miss the comment tagging system: insightful, informative, interesting, funny. It would make sense for hn.
15.
by benob 5mo ago
Space station tracking: https://flight-viz.com/cockpit.html?lat=40.64&lon=-73.78&alt...
16.
by benob 5mo ago
I just realized that a hash function is nothing less than the output of a deterministic random number generator xored with some data
17.
by benob 5mo ago
No, the failure is the human written prompt
18.
by benob 6mo ago
The author emphasizes accessibility and coherence as a benefit but another interesting one is composability which does not emerge naturally in the world of UI. Create a UI for a pair of websites like a command line for grep and wc. LLMs alr
19.
by benob 6mo ago
"That allows us to license the open source project under the more permissive MIT license."
20.
by benob 6mo ago
I would say: - decomposition: discover a more general form of Fourrier transform to untangle the underlying factors - memorization: some patterns are recurrent in many domains such as power low - multitask: exploit cross-domain connections
21.
by benob 6mo ago
Ollama is a user-friendly UI for LLM inference. It is powered by llama.cpp (or a fork of it) which is more power-user oriented and requires command-line wrangling. GGML is the math library behind llama.cpp and GGUF is the associated file fo
22.
by benob 6mo ago
Maybe they quantized a bit too much the model parameters...
23.
by benob 6mo ago
This is the worst lay-people explanation of an AI component I have seen in a long time. It doesn't even seem AI generated.
24.
by benob 6mo ago
This reminds me of Intel talking about faster web browsing with the new Pentium
25.
by benob 6mo ago
The real question is when will you resort to bots for rejecting low-quality PRs, and when will contributing bots generate prompt injections to fool your bots into merging their PRs?
26.
by benob 6mo ago
Reminds me of "Universal pre-training by iterated random computation" https://arxiv.org/pdf/2506.20057 , with bit less formal approach. I wonder if there is a closed-form solution for those kinds of initializa
27.
by benob 6mo ago
Time to start zig++
28.
by benob 7mo ago
It's funny that real value is now in test suites. Or maybe it's always been...
29.
by benob 7mo ago
I don't think this would qualify as clean room (the Library was involved in learning to generate programs as a whole). However, it should be possible to remove the library from the OLMO training data and retrain it from scratch. But wh
30.
by benob 7mo ago
What about doing that with movies and music?

More ›