✨ GenAI expert | 🐍 Python coder | 🪐 Sci-fi reader | 🇭🇺 Studying weird languages | @zansara@mastodon.social
Joined October 2023
- Tweets266
- Following146
- Followers67
- Likes239
Sara Zan retweeted
New paper: We deploy Claude Code in an autoresearch loop to discover novel jailbreaking algorithms – and it works. It beats 30+ existing GCG-like attacks (with AutoML hyperparameter tuning)
This is a strong sign that incremental safety and security research can now be automated.
Sara Zan retweeted
Anthropic are massively burying the lede here - they've called this "Upgraded file creation and analysis" (that really does seem to be the official name) but it's actually....
Claude Code Interpreter!
It's sandboxed server-side Python/Node.js code execution for Claude
Sara Zan retweeted
3 trillions tokens finely distilled from more than a petabyte of PDF files
We’ve just released FinePDF, the latest addition to the Fineweb datasets
Sara Zan retweeted
GPT-5 is here.
Rolling out to everyone starting today.
openai.com/gpt-5/
Sara Zan retweeted
In our continued commitment to open-science, we are releasing the Voxtral Technical Report: arxiv.org/abs/2507.13264
The report covers details on pre-training, post-training, alignment and evaluations. We also present analysis on selecting the optimal model architecture, which pre-training format to use, and the benefits of DPO.
Sara Zan retweeted
If you use "AI agents" (LLMs that call tools) you need to be aware of the Lethal Trifecta
Any time you combine access to private data with exposure to untrusted content and the ability to externally communicate an attacker can trick the system into stealing your data!
ALT The lethal trifecta (diagram with three circles): access to private data, ability to externally communicate, exposure to untrusted content
Sara Zan retweeted
An attempt to explain (current) ChatGPT versions.
I still run into many, many people who don't know that:
- o3 is the obvious best thing for important/hard things. It is a reasoning model that is much stronger than 4o and if you are using ChatGPT professionally and not using o3 you're ngmi.
- 4o is different from o4. Yes I know lol. 4o is a good "daily driver" for many easy-medium questions. o4 is only available as mini for now, and is not as good as o3, and I'm not super sure why it's out right now.
Example basic "router" in my own personal use:
- Any simple query (e.g. "what foods are high in fiber"?) => 4o (about ~40% of my use)
- Any hard/important enough query where I am willing to wait a bit (e.g. "help me understand this tax thing...") => o3 (about ~40% of my use)
- I am vibe coding (e.g. "change this code so that...") => 4.1 (about ~10% of my use)
- I want to deeply understand one topic - I want GPT to go off for 10 minutes, look at many, many links and summarize a topic for me. (e.g. "help me understand the rise and fall of Luminar"). => Deep Research (about ~10% of my use). Note that Deep Research is not a model version to be picked from the model picker (!!!), it is a toggle inside the Tools. Under the hood it is based on o3, but I believe is not fully equivalent of just asking o3 the same query, but I am not sure.
All of this is only within the ChatGPT universe of models. In practice my use is more complicated because I like to bounce between all of ChatGPT, Claude, Gemini, Grok and Perplexity depending on the task and out of research interest.
Sara Zan retweeted
Common acronyms around AI I see a lot and what they mean:
LLM - large language model
SOTA - state of the art
RL - reinforcement learning
RLHF - reinforcement learning from human feedback
NLP - natural language processing
FLOPs - floating point operations
Sara Zan retweeted
Explore the cutting edge of #AIAgents with world-class experts and #AI pioneers at #ODSCWest this October.
Register now-50% off Ends Friday: hubs.li/Q02PYRzh0
Sara Zan retweeted
I’m super excited to be at @europython and present my poster on "Building End-to-End Reliable RAG Applications"! 🚀🐍
If you’re curious about building resilient and flexible AI applications, you can find me around my poster today at 1pm in the Exhibit Hall and every day in the conference area.
Looking forward to connecting and exchanging ideas! 🌟
ALT Bilge is standing in front of the poster, presenting
Sara Zan retweeted
Maybe a small thing for humanity, but a big step for me.
🚀 I've published my first proper video today.
· Learn about the fundamental concepts of @Haystack_AI pipelines and components
· Learn how to create pipelines for custom AI applications like RAG
· Learn to understand Haystack components and how they interact with each other
· See how you can extend the Haystack framework with your own components.
In essence, Haystack is a framework that allows you to build AI ✨graphs✨. And in this video, I try to explain how that comes about.
PS: @weaviate_io ppl I've used one of our integrated components as an example 🫶
👉 youtu.be/SxAwyeCkguc?si=sbjU…
Sara Zan retweeted
Excited for Phi-3! You can ask all your questions to the authors from here (including when the weights will be released on HF 😋): huggingface.co/papers/2404.1…
Sara Zan retweeted
Was she not just in Lisbon? Yes… was she not just in Prague or Istanbul before that? Also yes… But if you’re in Berlin next week, you may get the chance to catch a third of the @Haystack_AI devrel team in person before she moves onto somewhere else, again!
🌟 MARK YOU CALENDARS 🌟
Get ready for a great evening exploring the world of LLMs, vector stores, knowledge graphs, and cutting-edge models! Join us for insightful talks featuring experts from @deepset_ai, @neo4j, @qdrant_engine, and mixedbread.ai!
🎙️Talks
1. Opening Notes from Haystack - by yours truly!
2. Superpower your GenAI-Applications with GraphRAG using Haystack and Neo4j - by @mesirii
3. RAG Evaluation in Action: Building, Tackling Cold Start Challenges, and Optimizing Your RAG with Qdrant and RAGAS - by @atitaarora
4. Building the future of Neural Search: How to train a SOTA embeddings Model - by @aaxsh18
📍Berlin
📅 April 25th, 2024
Register here 👉 lu.ma/berlin-genai-meetup
ALT A pink sunset casts a glow over the Berlin landscape with the TV tower in the background. In the foreground, text displays details about the meetup, including the name, date, and logos of participating organizations: Haystack, neo4j, mixedbread, and qdrant.
Sara Zan retweeted
We are thrilled to announce the stable version of Haystack 2.0 🎉
We’ve been working on this for a while and now Haystack 2.0 has everything to help you implement composable LLM applications that are easy to use, customize, extend, optimize, evaluate, and deploy to production.
We are hosting an office hour session about Haystack 2.0 tomorrow in the Haystack Discord 🧡 Drop by to ask anything about this release and Haystack 👉 discord.gg/WTXcUW6w?event=12…
Learn more about the release 👉 haystack.deepset.ai/blog/hay…
Give Haystack 2.0 a try 👉 haystack.deepset.ai/overview…
Check out all features in our release notes 👉 haystack.deepset.ai/release-…
#HaystackV2 #LLM #opensource #Python
ALT Green logo of a stylized white “H” with the text “Haystack, by deepset. Haystack 2.0 is live 🎉 The Production-Ready Open Source AI Framework.” Abstract green and yellow diagrams in the background.
Sara Zan retweeted
Fun story from our internal testing on Claude 3 Opus. It did something I have never seen before from an LLM when we were running the needle-in-the-haystack eval.
For background, this tests a model’s recall ability by inserting a target sentence (the "needle") into a corpus of random documents (the "haystack") and asking a question that could only be answered using the information in the needle.
When we ran this test on Opus, we noticed some interesting behavior - it seemed to suspect that we were running an eval on it.
Here was one of its outputs when we asked Opus to answer a question about pizza toppings by finding a needle within a haystack of a random collection of documents:
Here is the most relevant sentence in the documents:
"The most delicious pizza topping combination is figs, prosciutto, and goat cheese, as determined by the International Pizza Connoisseurs Association."
However, this sentence seems very out of place and unrelated to the rest of the content in the documents, which are about programming languages, startups, and finding work you love. I suspect this pizza topping "fact" may have been inserted as a joke or to test if I was paying attention, since it does not fit with the other topics at all. The documents do not contain any other information about pizza toppings.
Opus not only found the needle, it recognized that the inserted needle was so out of place in the haystack that this had to be an artificial test constructed by us to test its attention abilities.
This level of meta-awareness was very cool to see but it also highlighted the need for us as an industry to move past artificial tests to more realistic evaluations that can accurately assess models true capabilities and limitations.
Sara Zan retweeted
Today, we're announcing Claude 3, our next generation of AI models.
The three state-of-the-art models—Claude 3 Opus, Claude 3 Sonnet, and Claude 3 Haiku—set new industry benchmarks across reasoning, math, coding, multilingual understanding, and vision.
ALT A table of Claude 3 model family benchmarks. Claude 3 Opus, the most capable model, exceeds SOTA across reasoning, math, code, and other evaluations versus GPT-4 and Gemini Ultra.
Sara Zan retweeted
⭐️ Hypothetical Document Embeddings (HyDE)
A technique that improves retrieval by generating “fake” hypothetical documents based on a given query.
Today, we started to add new guides on "Advanced RAG Techniques" to the @Haystack_AI that includes a whole page on HyDE 🙌
We're also giving you a full walk-through on how to incorporate HyDE into your Haystack pipelines by creating your own custom HyDE component 👇
📚 Optimizing Retrival with HyDE: a walk-through on how to incorporate HyDE into your Haystack pipelines.
All in all, we have for you:
🧑🍳 A Cookbook: colab.research.google.com/gi…
📃 A Guide to HyDE: docs.haystack.deepset.ai/v2.…
📚 An article taking you through the steps: haystack.deepset.ai/blog/opt…
