47 karma · joined November 29, 2016
Before founding Mem0, I worked at Tesla as a Senior Autopilot Engineer, where I led the development of the AI Platform. This platform was essential for large-scale training and model evaluation for the Autopilot team. My role involved monitoring job performance, troubleshooting training clusters, automating machine learning workflows, and developing a centralized platform for engineers and researchers to access metrics, logs, and data.
My experience also includes creating various tools and infrastructure to improve machine learning processes, including 3D labeling software, model evaluation tools, and active learning-based label enhancement software.
Academically, I hold a Master's degree in Computer Science with a focus on Machine Learning from Georgia Tech. My research has concentrated on enhancing the accuracy of Artificial Intelligence and understanding human intentions and behaviors. I've been honored with the Snap Research Scholarship and gained experience as an intern at Snap Research.
Additionally, I spearhead EvalAI, an open-source platform dedicated to evaluating and comparing ML and AI algorithms at scale.
Shameless plug: We have been working on this problem at Mem0 to solve the long-term memory problem with LLMs. GitHub: https://github.com/mem0ai/mem0
1. LLM Compatibility: Mem0 works with various AI providers (OpenAI, Anthropic, Groq, etc.), while ChatGPT memory is tied to OpenAI's models only.
2. Target Audience: Mem0 is built for developers creating AI applications, whereas ChatGPT memory is for ChatGPT users.
3. Quality and Performance: Our evaluations show Mem0 outperforms ChatGPT memory in several areas:
- Consistency: Mem0 updates memories more reliably across multiple instances.
- Reliability: ChatGPT memory can be inconsistent with the same prompts, while Mem0 aims for more consistent performance.
- Speed: Mem0 typically creates memories in about 2 seconds, compared to ChatGPT's 30-40 seconds to reflect new memories.
These differences make Mem0 a better choice for developers building AI apps that need efficient memory capabilities.
- Inclusion prompt: User's travel preferences and food choices - Exclusion prompt: Credit card details, passport number, SSN etc.
Although we definitely think that there is scope to make it better and we are actively working on it. Please let us know if you have feedback/suggestions. Thanks!
- Control over what to remember/forget - Ability to set how detailed memories should be (some want more detailed vs less detailed) - Different structure of the memories based on the use case
Although I am curious about if there is enough market for this product. Would love to hear back any market research you guys have done before. Thanks! :)
I am Deshraj, a co-founder at Caliper, a platform for recruiting AI talent.
As ML/AI roles grow, we need scalable ways to test candidates' practical ML skills at the screening stage (before they come for an onsite interview).
This exists for testing CS algorithms/data structures in the usual coding interviews, but performance on those isn't necessarily correlated with practical ML / AI skills.
It would be nice if recruiters or hiring managers could create a challenge around some (say toy) data, invite candidates to participate in the challenge, see how they compare on a leaderboard, look at their code if they'd like, and then decide if they'd like to interview (and eventually hire!) the candidates.
Caliper provides exactly this. Would this be of interest?