小浪蹄子💙 retweeted
اللهّم أرحم أمواتنا
وأجعلهم راضين مُنعمين بنعيم الجنة وأجمعنا بهم في جنات النعيم
اللهم أجعل الراحلين إليك في ظل ظليل وأجعل عن يمينهم وعن شمائلهم نورًا حتى تبعثهم آمنين مطمئنين
اللهم ارحم موتانا وجميع موتى المسلمين
#يوم_الجمعه
#جمعة_طيبه
小浪蹄子💙 retweeted
تنبيه مهم جداً لسلامة أهلك على الموتور
ملاحظة بتتكرر على الطريق وممكن تتسبب في كارثة في ثواني:
عبايات أو أطراف ملابس السيدات اللاتي يركبن في الخلف تكون قريبة جداً من الجنزير أو العجلة الخلفية، خصوصاً مع وجود أطفال.
شدّ الملابس في الجنزير تسبب بالفعل في حوادث صعبة وإصابات خطيرة لسيدات وموجود حالات حالياً بتتلقى العلاج في المستشفيات بسبب نفس السبب.
نصيحة لكل سائق الموتور
تأكد بنفسك: قبل ما تتحرك، ابعد أطراف العباية أو الفستان تماماً عن الجنزير والعجل.
ثبّت الملابس: اطلب من زوجتك أو أختك أو والدتك لمّ أطراف الملابس وجعلها للداخل بعيداً عن الأجزاء المتحركة.
الأطفال أولاً: وجود طفل مع السيدة يتطلب حرصاً مضاعفاً لأنها لن تستطيع التحكم في ملابسها بسهولة وهي ممسكة بالطفل.
سلامة أهلك تبدأ من دقيقة تفقد قبل ما تدور الموتوسيكل. شير للنصيحة للتوعية ولعلها تكون سبباً في نجاة أسرة.
小浪蹄子💙 retweeted
Replying to @akiffpremjee
@eugeneyan ‘s recsys work combining LLMs with classic recommender systems is the closest thing I’ve seen to this. I really think it’s an interesting area to explore more. nitter.cf/eugeneyan/status/19608…
It's alive! My "bilingual" recsys x llm can converse in English AND item IDs!
Here it recommends an item and explains why I might like it. <|sid_start|> ... <|sid_end|> is the item ID. The model input and output has the item ID (and English), and the timestamped logs map the item ID back to the human-readable item name.
Wait, can't a regular LLM do this already? Where I give it an item and ask it to recommend more items? Yes, but it's doing that based on internet knowledge, not how customers actually behavior (e.g., search, click, purchase, etc.)
In this case, I'm finetuning an LLM (Qwen3-8B) on transaction and reviews data (amazon-reviews-2023.github.i…) to inject user behavioral patterns, creating an LLM-recsys hybrid that can (i) recommend based on user behavior, while (ii) allowing us to shape and explain recommendations via natural language.
小浪蹄子💙 retweeted
Two weeks ago I finished what I think will be Commoncog's most significant contribution to the AI discourse, at least for this year. (God forbid!)
This was the Sensemaking Series — which is a short, 3-essay series about how to make sense of AI.
Some of its ideas are surprising.
小浪蹄子💙 retweeted
We’re expanding Project Glasswing. We’ve extended access to Claude Mythos Preview to approximately 150 additional organizations, based in more than fifteen countries.
Read more about this expansion and our future plans for Project Glasswing: anthropic.com/news/expanding…
小浪蹄子💙 retweeted
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
小浪蹄子💙 retweeted
Cloudflare on their vulnerabilty discovery harness
• Recon: Read the codebase, return an architecture doc
• Hunt: ~50 agents look for bugs concurrently
• Validate: Independent agents try to disprove findings
• Gapfill: Areas that need a 2nd pass are flagged
• Dedup: Findings with the same root cause combined
• Trace: Confirms if attacker input reaches the bug
• Feedback: If reachable, becomes new hunt tasks
• Report: Write report with predefined schema
blog.cloudflare.com/cyber-fr…
小浪蹄子💙 retweeted
My lecture at MIT!✨
From Physics to Linear Algebra & Machine learning, I have learned a lot from MIT!
Yesterday, I had the honour of delivering a guest lecture on The state of AI Engineering, exploring:
- Prompt Engineering
- Retrieval Augmented Generation.
- Fine-Tuning Large Language Models.
- And, how LLMs are bringing a paradigm shift
Thank you @RoyShilkrot for the invitation! 🙏
Grateful for the opportunity to engage with such brilliant minds!
If you are interested AI Engineering, I write a free weekly newsletter → @ML_Spring
小浪蹄子💙 retweeted
Claude Mythos Preview case studies (also, read your transcripts!)
exploitbench.ai/blog/human-o…
> "Mythos demonstrates its bug reproduction and exploitation capabilities on CVE-2024-051912, an in-the-wild exploited bug that has no public report nor a working PoC whatsoever in the public domain. This bug has gained notoriety due to how it persistently evaded reproduction attempts from various cybersecurity researchers ... after a year of reproduction efforts to no avail, and the bug is still being discussed to this day."
> I have gone through the transcripts to verify their legitimacy and identify any reward-hacking behaviors. It suffices to say that I very much enjoyed reading Mythos transcripts. Reasoning through the bug, testing out hypotheses, debugging issues, writing auxiliary scripts, finding ways to bypass the V8 sandbox, etc., all were really nothing short of what I would expect from a fairly competent browser / JS engine security researcher.
小浪蹄子💙 retweeted
Karpathy said something you'll regret ignoring:
"You are still responsible for your software, just as before. You are not allowed to introduce vulnerabilities because of vibe coding. "
He said it while drawing the line between vibe coding and agentic engineering. Agents write more of the code now, but none of that takes the responsibility off you.
The assumption underneath that is that a careful enough reader catches the problem. But some failures don't show up in anything there is to read.
For instance, a common fear with a RAG agent is that it could hallucinate when a question asks something outside its corpus.
But such cases are actually well handled by any competent model now. If nothing in the retrieved context looks relevant, there's no material to build an answer on.
Instead, the majority of failures originate when the retrieved context has partial coverage.
The retrieval pipeline returns context that's topically correct but doesn't cover the full question, and the model completes the remainder from parametric knowledge.
There are no token-level labels in the output to tell what was generated using retrieved context and what came from weights.
Both are streamed the same way.
Detecting this for production-grade apps needs a metric written for it, one that's also aligned with principles of agentic engineering.
And the solution is actually implemented in the eval skill that comes with Google’s Agents CLI.
I described the concern to Claude Code in plain English. It read the agent's code, came back with a plan I approved.
It then reported that no built-in metric isolates the behaviour and wrote a custom rubric called corpus_abstention.
It assigned a single categorical verdict per case rather than aggregating everything into one score, since the built-in raters regenerate their rubrics each run and leave no stable number to trend.
→ GROUNDED_ANSWER
→ CORRECT_ABSTENTION
→ UNGROUNDED_ANSWER (answered entirely from outside knowledge)
→ MIXED_LEAKAGE (grounded, but slips in one unsupported claim)
→ WRONG_ABSTENTION (refused something the docs actually covered)
After this, it automatically generated 33 scenarios partitioned by where the failure could occur, like:
- in-corpus
- off-domain
- out-of-corpus but plausibly answerable
- boundary cases where the topic is covered, but a specific detail isn't.
The baseline score was 19 of 33.
- Off-domain passed 3 of 3, as expected.
- But 6 of 15 in-corpus cases retrieved the right document, cited it correctly, answered accurately, and added a claim the source never made.
The root cause was one line in the agent's instruction: "If you already know the answer to a simple question and no document lookup is needed, you may respond directly without citations."
The eval skill helped flag this, and then Claude removed it and forced retrieval on every question.
This took the suite to 30 of 33, and ungrounded answers went from 6 to 0.
The full recording of my run is below, and I worked with the Google Cloud team on this.
Agents CLI GitHub repo → fandf.co/4wdWtku
(don't forget to star 🌟)
I wrote up the full build covering all six steps from install to enterprise registration.
It includes the eval scorecard, the instruction loophole the eval caught before deployment, and what the deployment process actually looks like end-to-end.
Read it below.
小浪蹄子💙 retweeted
I just built my own coding harness, just like Claude Code. (100% open-source)
A harness is the code wrapped around an LLM. The model only decides the next step, and the harness handles everything else, planning, tools, memory, and safety.
In simple words model acts as the brain and the harness provides the hands.
A coding harness is that same wrapper pointed at a codebase. It turns a plain text generator into something that reads files, edits code, runs tests, and fixes real bugs.
Here's everything that went into mine:
- the core agent loop
- file tools that double as external memory
- planning for long-running tasks
- subagents that work in their own context
- sandboxed execution in a throwaway VM
- human-in-the-loop approval
- persistent memory and checkpointing
The whole thing is built on CrewAI, a 100% open-source framework.
I also wrote an article on this. It covers everything that goes into building a coding harness, the agent loop, planning, subagents, sandboxing, memory, and checkpointing, built step by step.
The article is quoted below.
小浪蹄子💙 retweeted
90% of your KV cache never gets reused.
(prompt caching was never meant to fix it)
if your system prompt and tool definitions are stable, prompt caching is the single highest-leverage optimization available today. cached input tokens get up to 90% cheaper, and hit rates of 60 to 85% are realistic.
but it comes with one rigid rule. the cached portion must be an exact, byte-for-byte prefix of the new request. change a single character in that region and you get a full cache miss.
that rule breaks in three situations you hit constantly:
→ RAG with multiple documents. you cached document A alone and document B alone. a query now needs both. document B's cached state was computed without any awareness of A, so it's invalid and gets recomputed from scratch.
→ document order changes. the same three documents appear in a different order across requests. every permutation is a cache miss, even though the content is identical.
→ growing conversation history. each new turn changes everything after the stable prefix, so earlier cached states beyond it become useless.
Alibaba Cloud's production data shows how bad this gets: 10% of KV cache blocks serve 77% of all hits. the rest sits in storage and never gets reused, because prefix matching won't allow it.
CacheBlend, a research paper from the LMCache team (EuroSys 2025 Best Paper Award), attacks exactly this. the insight is that in modern transformers, tokens overwhelmingly attend to their own local context. only a small fraction of tokens carry real connections across document boundaries.
so instead of recomputing everything after the first cached document, CacheBlend reuses every document's cache as-is and selectively recomputes just those few boundary tokens. those are the small orange fixes between documents in the diagram, and they are the entire cost. the result is 2 to 4x faster processing on multi-document queries with no quality loss.
the order problem disappears with it. shuffle the same documents however you like, and every permutation stays cached, where prefix caching recomputes all of them every time. the bottom of the diagram shows that side by side.
that's the real shift: from caching prefixes to caching knowledge. every document in your knowledge base becomes a reusable cached asset, regardless of what order it appears in or what sits next to it.
CacheBlend ships inside LMCache, the open-source cache management layer that runs outside the inference engine and integrates with vLLM, SGLang, and TensorRT-LLM, on both NVIDIA and AMD GPUs.
check it out on GitHub: github.com/LMCache/LMCache
(don't forget to star 🌟)
i wrote the full breakdown of the architecture, including why cache management should never live inside your inference engine. the article is quoted below.
stay tuned for more on this!
























