Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Agents need continual learning. At supermemory, we are doubling down and pushing the frontier for memory and in-context learning, for every agent and use case. Introducing 𝚕𝚎𝚊𝚛𝚗𝚎𝚛-𝟷

1,869,281 Aufrufe • vor 2 Monaten •via X (Twitter)

34 Kommentare

Profilbild von Adavya Sharma
Adavya Sharmavor 2 Monaten

so what actually is learner-1

Profilbild von supermemory
supermemoryvor 2 Monaten

You can use supermemory to power your agents, today:

Profilbild von Kaan Demirel
Kaan Demirelvor 2 Monaten

if the "learning" is just dynamically injecting better tokens into ctx, that’s still retrieval imo. interested to see what learner-1 is actually changing

Profilbild von Joaquin Bonifacino
Joaquin Bonifacinovor 2 Monaten

This is vague af, "continual learning" okey awesome "injecting tokens in context memory", it is markdown memory files again is it?, please...

Profilbild von Krish Jaiswal
Krish Jaiswalvor 2 Monaten

this is confusing. injecting tokens into model's context in real time ain't continual learning. how do you actually do it then? and ICL has been there since the inception of agent memory field.

Profilbild von Karl-Gustav Kallasmaa 🔍
Karl-Gustav Kallasmaa 🔍vor 2 Monaten

Great work. You're finally learn-1 something

Profilbild von Aayush
Aayushvor 2 Monaten

Cool stuff, but I think you guys need a better microphone.

Profilbild von yam
yamvor 2 Monaten

What is "in-context" learning here? Do you mean continually updating the input fed into the LLM?

Profilbild von Ricardo Mendez
Ricardo Mendezvor 2 Monaten

When it comes to injecting tokens into the context windows, what is the difference between learner-1 and what supermemory already does?

Profilbild von Vishal Anton
Vishal Antonvor 2 Monaten

This is super cool. Congrats guys!!

Profilbild von Yakko
Yakkovor 2 Monaten

congrats!! looks awesome

Profilbild von ari dutilh
ari dutilhvor 2 Monaten

oh shit

Profilbild von Abhilaksh
Abhilakshvor 2 Monaten

Building something similar :)

Profilbild von Shrey Jindal
Shrey Jindalvor 2 Monaten

Sadly memory will not be solved by context injection or ranking algorithms, it'll be an active process turning traces into standardized guides. Take from this what you will ;)

Profilbild von John Hawkins
John Hawkinsvor 2 Monaten

Isn’t that what the supermemory plugin does ??

Profilbild von Yash Hulsurkar
Yash Hulsurkarvor 2 Monaten

wooooo!! congrats guys!

Profilbild von Subhash Yadav
Subhash Yadavvor 2 Monaten

Continual learning also means continual writing: every fact an agent carries into tomorrow was taught to it by something it read today. That makes memory the persistence layer for prompt injection - a poisoned 'fact' survives the session it arrived in. Memory writes need provenance and review, not just storage.

Profilbild von Harsh Savergaonkar
Harsh Savergaonkarvor 2 Monaten

this is amazing 🔥🔥

Profilbild von AI Mastery Guide
AI Mastery Guidevor 2 Monaten

ok learner-1 sounds interesting, what does it actually do

Profilbild von sans
sansvor 2 Monaten

holy larp

Profilbild von Vineeth
Vineethvor 2 Monaten

crazyyyyy stufff

Profilbild von Vatsalpandya333
Vatsalpandya333vor 2 Monaten

W

Profilbild von Singularity
Singularityvor 2 Monaten

Memory is the very core and basic part of any AI agents. Just like chips, just after the chips or hardware, memory is really important, and just after that, alignment comes.

Profilbild von UsefulScout | Tools for Builders
UsefulScout | Tools for Buildersvor 2 Monaten

The hard part after “agents can remember” is probably deciding what they should forget. Persistent memory gets much more useful when it has confidence, recency, deduping and expiry otherwise agents just accumulate stale context. Memory quality may matter more than memory size.

Profilbild von Jatin Garg
Jatin Gargvor 2 Monaten

I do this manually. Rewrite the rules after every failed session and the next one goes better. But the agent itself starts blank every time.

Profilbild von Fat'hah Noor Prawita
Fat'hah Noor Prawitavor 2 Monaten

@grok apa ini?

Profilbild von CryptoPraetoria
CryptoPraetoriavor 2 Monaten

Why is supermemory not working?

Profilbild von Fajar M Reza
Fajar M Rezavor 2 Monaten

Continual learning matters when agent memory improves decisions without bloating prompts.

Profilbild von Karthik Varma
Karthik Varmavor 2 Monaten

Memory is useful, but agents learning from what happened last time is the real leap. How does learner-1 handle conflicting lessons over time?

Profilbild von The Black Box
The Black Boxvor 2 Monaten

continual learning is the only way agents stop forgetting yesterday's context. practical memory beats flashy in-context tricks every time

Profilbild von rawcash
rawcashvor 2 Monaten

is it really continual learning tho??

Profilbild von ZIL
ZILvor 2 Monaten

Just like we learn from our daily mistakes, learner-1 helps agents grow, and I am so proud of you @0xndra for making this happen.

Profilbild von _alphashark_
_alphashark_vor 2 Monaten

Continual learning's hardest failure mode isn't catastrophic forgetting, it's your agent confidently acting on a memory that was accurate 3 weeks ago. Staleness detection never makes it into the agent demos.

Profilbild von Purav
Puravvor 2 Monaten

continual learning is a meaningless term, especially in this context. but hey I guess marketing hype helps you shill your product.

Ähnliche Videos

Models have IQ but no experience. Frontier models show up on day one with superhuman intelligence but no expertise on the thousands of tasks your company actually does. Trajectory exists to close that gap through continual learning. Ronak Malde trained SWE-1 at Windsurf, then went to DeepMind. His cofounders Michael Elabd and Arjun Karanam worked on Gemini Robotics and multimodal models for Vision Pro. All three hit the same wall: big labs train one model for everyone, but nobody trains one for you. The conversation covers: — why today’s agents are “pseudo continual learning”: we are in the paper road map era, not Google Maps — “Group Size 1”, a GRPO variant that learns from one real-world rollout instead of replaying the same task eight times, so you don’t need a simulated environment — how Harvey trained a model 6x cheaper than frontier that took an open-source baseline up by over 4x on its own legal benchmark — Continual Grading, a meta-grader that turns rubrics on 100 traces into reward for 10,000 — continual learning vs. recursive self-improvement, and why only one keeps humans in the loop — why continual learning will make product-building more like tending a garden of products Chapters 0:00 – Introduction 0:45 – Pseudo continual learning: paper maps vs Google Maps 2:48 – Do we need new architectures? Rich Sutton, David Silver, and the missing layer 4:33 – From a $100M consulting model to a product: lessons from SWE-1 at Windsurf 7:36 – Harvey: 6x cheaper than frontier, 4% to 18% on its own benchmark 8:55 – Why "Trajectory": the unit of work is a task, not a token 10:13 – The primitives of continual learning: traces, replayable environments, reward 12:56 – Group Size 1 and the meta-grader: training without an environment or a labeled reward 16:16 – The platform: train, eval and deploy in one click, and who it's for 18:58 – Rogo and intelligence density 20:14 – What's next: curiosity, and why continual learning beats RSI on alignment 24:19 – Research and product under one roof, and the product builder as gardener 27:26 – Why work at Trajectory, and the optimistic case: from 1% to 99%

Lauren Reeder

36,750 Aufrufe • vor 11 Tagen