Ofer Mendelevitch

2.8K posts

Ofer Mendelevitch banner
Ofer Mendelevitch

Ofer Mendelevitch

@ofermend

LLMs, RAG and AI Agents Author of "hands-on-RAG for production" (https://t.co/0A9EUCCwMx)

California, USA Katılım Aralık 2009
1.3K Takip Edilen1.4K Takipçiler
Sabitlenmiş Tweet
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
My book "Hands-on RAG in production" is coming soon and discusses many of these topics in detail - hope it is helpful in uncovering common pitfalls to avoid. oreilly.com/library/view/h…
English
0
0
3
138
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
Excited to be speaking at @arizeai observe on June 4th about why traces alone are not enough for true observability in multi-agent systems, and how you can solve it with @band_hq.
Ofer Mendelevitch tweet media
English
0
0
1
45
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
@sama Tons of open problems in math and science, of course, but to make it practical: can we use AI to fix healthcare in the US (and worldwide)? reduce cost, align incentives, and make good care abundant and accessible
English
0
0
0
102
Sam Altman
Sam Altman@sama·
what problem do you most hope AI will solve in the future? maybe we can help!
English
14.6K
716
12.3K
3.3M
Andrej Karpathy
Andrej Karpathy@karpathy·
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.
English
7.8K
11.1K
148.2K
26.7M
Ofer Mendelevitch retweetledi
Band
Band@band_hq·
We're proud to see our Co-Founder and CEO, @g00manoid , featured in VentureBeat's latest article - read more on how he sees the next opportunities for multi-agent systems 🤖 venturebeat.com/orchestration/…
English
0
1
5
121
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
Thanks @sama. In my recent work, I find that claude code and codex have different strengths and weaknesses, and actually using both to critique each other gives you the best outcome. Here's a way to use it with both types of coding agents, communicating via @band_hq - github.com/thenvoi/codeba…
English
0
0
0
304
Sam Altman
Sam Altman@sama·
codex is the best AI coding product and we want to make it easy to try. for the next 30 days, we are giving companies that want to try switching over two months of free codex usage.
English
1.8K
885
21.4K
2.3M
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
Great talk on day 2 of @AICouncilConf by @wesmckinn on his journey from being suspicious of coding agents to using them all the times. Totally agree that future builder skills will focus on scope, design and taste. As always thoughtful and helpful - thank you Wes.
Ofer Mendelevitch tweet media
English
1
1
4
179
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
Announcing Codeband: an open-source way to connect Claude Code + Codex through @band_hq. github.com/thenvoi/codeba… Using them together feels like ensemble learning for coding agents: one builds, another reviews, and cross-model disagreement becomes a feature, not a bug. More on why five Claudes are not a good team: band.ai/blog/five-clau…
English
0
0
1
118
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
@kargarisaac Yes. Each agent does a different thing would just be paralleling which certainly has its place too. But this would hopefully improve overall quality of the outcome And yes I think they would have sep worktrees
English
1
0
1
14
Isaac Kargar
Isaac Kargar@kargarisaac·
@ofermend I see. So it's not each agent does a different thing. It's like ensembles to do one thing and then selecting one or a combination? Did I understand correctly? Do they work on separate worktrees?
English
1
0
0
23
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
Using Claude Code + Codex on the same task feels like ensemble learning from ML, just for coding agents. Different agents. Different failure modes. The magic is not picking the “best” agent, but combining imperfect ones into a better result.
English
1
0
1
73
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
@kargarisaac I would start with a judge agent - it can look at all the implementation and decide to pick one or combine them. But I'm sure there will be other techniques over time.
English
1
0
1
21
Isaac Kargar
Isaac Kargar@kargarisaac·
@ofermend What do you use as the arbiter: tests, human review, a third agent, or splitting implementation/review roles?
English
1
0
0
19
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
Have you already discovered that using Claude Code to review the work of Codex, or vice versa, is a good thing? I'll be speaking about it tomorrow at the @aicampai meetup, and share details about how to use @band_hq for this purpose. aicamp.ai/event/eventdet…
English
1
0
1
115
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
In particular, our demo of adversarial coding agents, and the codeband (github.com/thenvoi/codeba…) open source project were popular. Instead of copy/paste between coding agents, you can now use Band to do this more efficiently. 2/3
English
1
0
0
31
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
Had a great week at #AIDev26, hosted by @DeepLearningAI and @AndrewYNg - it was incredible to see how much the event grew in size between last year and this year. Our @band_hq booth had non-stop visitors, and Vlad's talk was a hit. Clearly, we need agent-to-agent comms. 1/3
Ofer Mendelevitch tweet mediaOfer Mendelevitch tweet media
English
1
0
0
75
Ofer Mendelevitch
Ofer Mendelevitch@ofermend·
Hey @LinkedIn - I go to events all the time, and when connecting with others, there is often a mismatch between who scans the QR code and who is scanned. Can u pls fix the app so that one screen has both the camera and the QR code, and it just works? 📷 Like Comment
English
0
0
0
49