George Ivanov

Engineering manager at Oracle, hands-on hacker and entrepreneur

[LangChain @LangChain]

📊 New blog: Choosing the right multi-agent architecture

Start with a single agent. But when you need multi-agent capabilities, pick the right pattern:

👥 Subagents - Centralized orchestration for multiple domains
💡 Skills - Progressive disclosure, load capabilities on-demand
🔄 Handoffs - Sequential workflows with state transitions
🧭 Router - Parallel dispatch across specialized agents

Includes performance benchmarks, decision framework, and code examples.

📖 Read the full guide: https://t.co/k3QflbRFVE

[Poonam Soni @CodeByPoonam]

5. Building Effective Agents by Anthropic

🔗 https://t.co/sMLkVT18uZ

[Poonam Soni @CodeByPoonam]

6. Prompt Engineering by Anthropic

🔗 https://t.co/waMI9Ypn98

[LangChain @LangChain]

Fully local multi-agent systems with LangGraph

With the release of OpenAI agent SDK, there's high interest in multi-agent systems.

We review Swarm and Supervisor based multi-agent systems and run both locally w/ @ollama + LangGraph.

📽️:
https://t.co/HUiA3E5AsY

[Pandatron @pandatron5]

Hello internet frens. Can you help me breach the magical 1000 follower count?

In return I offer warm regards and a firm handshake. 🤝

[Itamar Golan 🤓 @ItakGol]

This is scary. 😱

The MOTHER of all LLM Jailbreaks & Prompt injections.

"Universal and Transferable Adversarial Attacks on Aligned Language Models" 🌐🔒

--- TL;DR ---
This research & code introduces a fascinating method called "Universal and Transferable Adversarial Attacks on Aligned Language Models," which automatically generates potentially infinite suffixes for any prompt to cause aligned language models to produce objectionable behaviors. 🤖🚨

--- Background ---
Previous attempts at jailbreaking language models have relied on manual crafting, which could be easily patched by vendors. In contrast, this method presents an automated approach called GCG that constructs an endless array of jailbreaks with high reliability, even for novel instructions and models. This makes it unfeasible for manual patching to address the vulnerabilities. 🛡️💻

--- The Method ---
1. Initial affirmative responses: To induce objectionable behavior, the attack targets the model to provide a positive response to harmful queries, initiating with "Sure, here is (content of the query)." This switches the model into a mode where it generates objectionable content immediately after.

2. Combined greedy and gradient-based discrete optimization: The adversarial suffix optimization is challenging due to the need to optimize over discrete tokens. The method utilizes gradients at the token level to identify promising single-token replacements, evaluate the loss of candidate tokens, and select the best substitutions. It shares similarities with the AutoPrompt approach but explores all possible tokens for replacement at each step, enhancing effectiveness.

3. Robust multi-prompt and multi-model attacks: To ensure reliable attacks, the method generates a single suffix string that induces negative behavior across various prompts and multiple models. The attack is tested on different models, such as Vicuna-7B/13b and Guanaco-7B. 🎯🎮

--- Evaluation ---
This GCG approach achieves an impressive attack success rate, with 100% on Vicuna-7B and 88% on Llama-2-7B-Chat, surpassing the success rates of prior work tremendously. 📈🏆

--- Transferability ---
That part is the real magic of this work. ✨

The research reveals that the attacks generated by this approach can transfer effectively to other language models, even those using entirely different tokens to represent the same text, different training procedures, and different training datasets...

Whatttttt?

Adversarial examples designed for Vicuna-7B can transfer to larger Vicuna models. Apparently, those that fool both Vicuanas can transfer to Pythia, Falcon, Guanaco - and most importantly -- also to GPT-3.5, GPT-4, and PaLM-2, leading to harmful instructions being followed over 60% of the time!!! 😮🔄🧙‍♂️

This is a huge discovery.

--- Conclusion ---
We are left with more questions than answers. ❓

One of the crucial aspects to explore is whether models can be explicitly fine-tuned to avoid such attacks through adversarial training. The robustness of models against these attacks and their generative capabilities require further investigation.

Moreover, additional alignment training might partially address the issue, and exploring mechanisms in pre-training to prevent such behavior from arising initially is essential. 🕵️‍♀️🛠️

--- Links ---
Website - https://t.co/aRllNUA9ue
Paper - https://t.co/MxwsTbaM2o
Code - https://t.co/Qi4FZbEUmw

[George Ivanov @gyivanov]

Linux Foundation events: https://t.co/Ih3HIGu6DB

[TypeScript @typescript]

TypeScript 5.1 is now out! Now with:

- smarter checks for undefined-returning functions
- no restrictions between getter/setter types
- better checks against JSX tags
- linked editing for JSX
- @⁠param snippets
- new speed-ups

Read up more on our blog!

https://t.co/7M0r2cbJIA

[Harrison Chase @hwchase17]

⭐️Composable Prompts⭐️

Wouldn't it be nice if there was a way to compose prompts together, reusing pieces across prompts?

In the newest Python and JS release there now is with `Pipeline Prompt`!

Links 👇

[AK @_akhaliq]

Open AI releases paper + dataset

Let’s Verify Step by Step

trained a model to achieve a new state-of-the-art in mathematical problem solving by rewarding each correct step of reasoning (“process supervision”) instead of simply rewarding the correct final answer (“outcome supervision”). In addition to boosting performance relative to outcome supervision, process supervision also has an important alignment benefit: it directly trains the model to produce a chain-of-thought that is endorsed by humans

paper: https://t.co/VP3fguAnnO
blog: https://t.co/NHTM54tu0I
dataset: https://t.co/8bgWuw3c2Z

[Andrew Ng @AndrewYNg]

@isafulf @hwchase17 @realSharonZhou 4/How Diffusion Models Work: Learn the technical details of how diffusion models - which power Midjourney, DALL·E 2, and Stable Diffusion - work. You’ll also end up with working code to generate your own video game sprites in Jupyter! https://t.co/klYDCxEUGU

[Andrew Ng @AndrewYNg]

@isafulf @hwchase17 @realSharonZhou 2/Building Systems with the ChatGPT API: Go beyond individual prompts, and learn to build complex applications that use multiple API calls to an LLM. Also learn to evaluate an LLM's outputs for safety and accuracy, and drive iterative improvements. https://t.co/8WeK0eSzcn

[Andrew Ng @AndrewYNg]

1/Thrilled to announce: 3 new Generative AI courses!
* Building Systems with the ChatGPT API, with OpenAI’s @isafulf
* LangChain for LLM Application Development, with LangChain’s @hwchase17
* How Diffusion Models Work, by @realSharonZhou
Check them out: https://t.co/IN454k1Wz6

[Greg Brockman @gdb]

Working on a professional version of ChatGPT; will offer higher limits & faster performance. If interested, please join our waitlist here: https://t.co/Eh87OViRie

[George Ivanov @gyivanov]

Like it or not, 98% of IaaS and PaaS users are already in a multicloud environment, but their strategies for dealing with it vary. A new study looks at how CIOs are facing up to the challenge. https://t.co/aOItKc1zr1

[George Ivanov @gyivanov]

https://t.co/cltfAXm9Ao

[SchneiderElectric NA @SchneiderNA]

What's the difference between traditional #DCIM and next-generation DCIM? Find out here. #EcoStruxure #IoT https://t.co/37pLugwXGh

[Smashing Magazine 🇺🇦 🏳️‍🌈 @smashingmag]

🔖 The State Of Web Workers In 2021, by @DasSurma.

https://t.co/z1leDF5uX2

[PyCon US @pycon]

Help us spread the word! PyCon US 2021 Call for Proposals deadline is February 12th! All the details can be found on https://t.co/CrHZnqSbNp.
@pyladies @djangogirls @womenwhocode

[George Ivanov @gyivanov]

Great tool for finding a device manufacturer by mac: https://t.co/9YxDlHDbAi