a2a-salesforce/a2-failures
Updated • 119 • 1
Sarath Shekkizhar
shekkizh
AI & ML interests
None yet
Recent Activity
updated a dataset about 11 hours ago
a2a-salesforce/a2-failures updated a dataset about 16 hours ago
a2a-salesforce/a2-bench published a dataset 1 day ago
a2a-salesforce/a2-failuresOrganizations
updated a
dataset about 11 hours ago
updated a
dataset about 16 hours ago
published a
dataset 1 day ago
Add competitive bilateral contracting scenarios in role-specific task-data format
1
#1 opened 4 days ago
by 
srijanbansal
Add clarified SLA-breach settlement task data
1
#2 opened 4 days ago
by 
pjwalapuram
Add competitive contracting runs: two-agent and four-agent experiments
1
#1 opened 4 days ago
by srijanbansal
upvoted a paper about 1 month ago
published a
dataset 3 months ago
authored a
paper 6 months ago
upvoted a paper 6 months ago
submitted a
paper to Daily Papers 6 months ago
authored a
paper 6 months ago
liked a
Space 7 months ago
What do we know about the architecture so far?
👍 3
5
#6 opened about 1 year ago
by 
amgadhasan
replied to their post over 1 year ago
Didn’t you know AGI is already here 🤖
replied to their post over 1 year ago
Images are split into patches and each patch is tokenized - the tokenization is taking into a feature dimension and quantizing. This is probably already has CNN and/or attention. The issue is that of the model not able to reason both color and text in the tokenized space.
We ran about 1000 experiments - different prompting, tool call to different model for recognition, and several other techniques. The results still hold. The paper is a small part of the analysis.🤷♂️
posted an update over 1 year ago
Post
2,503
🙋🏽♂️ Is your "multi agent" system really multi agentic? Or is it just a modular setup with a bunch of different prompts? 🤨
I’ve had this discussion way too often, so I finally wrote it all down. If you’re building with agents, you need to read this.
Here’s the TLDR:
✅ True multi agent systems require:
• Persistent, private state per agent
• Memory that impacts future decisions
• Adaptation based on past experiences
❌ Just having modular components, function calls, or multiple LLMs doesn't cut it. That’s not multi agentic. It’s just pipelining.
🤝 The magic is in evolving relationships, context retention, and behavioral shifts over time.
🧠 If your agents aren’t learning from each other or changing based on past experience… you are missing the point.
What do you think? Curious what patterns you're experimenting with 🧐
👉 Full post: https://shekkizh.github.io/posts/2025/04/multi-agents/
I’ve had this discussion way too often, so I finally wrote it all down. If you’re building with agents, you need to read this.
Here’s the TLDR:
✅ True multi agent systems require:
• Persistent, private state per agent
• Memory that impacts future decisions
• Adaptation based on past experiences
❌ Just having modular components, function calls, or multiple LLMs doesn't cut it. That’s not multi agentic. It’s just pipelining.
🤝 The magic is in evolving relationships, context retention, and behavioral shifts over time.
🧠 If your agents aren’t learning from each other or changing based on past experience… you are missing the point.
What do you think? Curious what patterns you're experimenting with 🧐
👉 Full post: https://shekkizh.github.io/posts/2025/04/multi-agents/
- 2 replies
posted an update over 1 year ago
Post
1,955
Think AGI is just around the corner? Not so fast.
When OpenAI released its Computer-Using Agent (CUA) API, I happened to be playing Wordle 🧩 and thought, why not see how the model handles it?
Spoiler: Wordle turned out to be a surprisingly effective benchmark.
So Romain Cosentino Ph.D. and I dug in and analyzed the results of several hundred runs.
🔑 Takeaways
1️⃣ Even the best computer-using models struggle with simple, context-dependent tasks.
2️⃣ Visual perception and reasoning remain major hurdles for multimodal agents.
3️⃣ Real-world use cases reveal significant gaps between hype and reality. Perception accuracy drops to near zero by the last turn 📉
🔗 Read our arxiv article for more details https://www.arxiv.org/abs/2504.15434
When OpenAI released its Computer-Using Agent (CUA) API, I happened to be playing Wordle 🧩 and thought, why not see how the model handles it?
Spoiler: Wordle turned out to be a surprisingly effective benchmark.
So Romain Cosentino Ph.D. and I dug in and analyzed the results of several hundred runs.
🔑 Takeaways
1️⃣ Even the best computer-using models struggle with simple, context-dependent tasks.
2️⃣ Visual perception and reasoning remain major hurdles for multimodal agents.
3️⃣ Real-world use cases reveal significant gaps between hype and reality. Perception accuracy drops to near zero by the last turn 📉
🔗 Read our arxiv article for more details https://www.arxiv.org/abs/2504.15434
- 3 replies
posted an update over 1 year ago
Post
624
Some interesting architectural choices made in Llama 4 models -- were these key to the 10M context? Possibly 🤔
🔍 Takeaways:
🧩 Interleaved Attention without position encoding
- LLaMA 4 removes explicit positional encoding in some attention layers to boost performance on longer contexts.
- The principles here could be similar to the residual connections to facilitate attention to early tokens without positional decay.
⚖️ Scaled Softmax to increase attention at inference time
- The max attention value (output of softmax) decreases as context size increases.
- Llama 4 incorporates a context-size dependent temperature in the softmax function to modify the slope of softmax, allowing the model to focus better on relevant tokens.
- Done only at inference time -- guessing it was more a choice after some observation on eval datasets.
What did you think of these choices?
🔍 Takeaways:
🧩 Interleaved Attention without position encoding
- LLaMA 4 removes explicit positional encoding in some attention layers to boost performance on longer contexts.
- The principles here could be similar to the residual connections to facilitate attention to early tokens without positional decay.
⚖️ Scaled Softmax to increase attention at inference time
- The max attention value (output of softmax) decreases as context size increases.
- Llama 4 incorporates a context-size dependent temperature in the softmax function to modify the slope of softmax, allowing the model to focus better on relevant tokens.
- Done only at inference time -- guessing it was more a choice after some observation on eval datasets.
What did you think of these choices?
updated a
Space over 2 years ago