Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

Meet the minds building supermemory! 🧠 Our research engineer Prasanna breaks down how memory actually works, in your head, and in AI. inside the episode: → why forgetting is a feature and not a failure → what your hippocampus inside brain is really doing all day → how "dreaming"... show more

supermemory

29,063 subscribers

14,752 просмотров • 2 месяцев назад •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

Комментарии: 8

Фото профиля MR ANDERSON

MR ANDERSON2 месяцев назад

@_prasanna_ap Really enjoyed the focus on how human memory and AI memory intersect. Looking forward to watching the full episode and learning from the team behind Supermemory. 🧠

Фото профиля Zohaib Ai

Zohaib Ai2 месяцев назад

@_prasanna_ap Bigger context windows improve recall, but they don't create memory. The real challenge is deciding what deserves to be remembered.

Фото профиля Hash

Hash2 месяцев назад

@_prasanna_ap If your product replicate human memory... it will fail miserably... because my brain cannot recall what I ate yesterday...! Also human memory is very bad at handling structured data...

Фото профиля NOOR TECH

NOOR TECH2 месяцев назад

@_prasanna_ap Meet the minds building @supermemory with @_prasanna_ap.

Фото профиля stanleyweb3

stanleyweb32 месяцев назад

@_prasanna_ap Nice to meet you sir

Фото профиля Aina Ai | Tools & Updates

Aina Ai | Tools & Updates2 месяцев назад

@_prasanna_ap Brilliant breakdown Forgetting is a feature and context memory This is the episode that actually makes AI brain science click

Фото профиля AqibAi

AqibAi2 месяцев назад

Memory isn’t about storing everything—it’s about knowing what matters. The gap between context windows and true memory is becoming one of the biggest challenges in AI. Understanding how humans forget, compress, and retain knowledge could shape the next generation of intelligent systems. Great conversation. 🧠

Фото профиля ZenithAi

ZenithAi2 месяцев назад

@_prasanna_ap A thoughtful discussion that makes AI memory far easier to understand

Похожие видео

Everyone wants AI with a bigger memory, we think that's the wrong goal.. We sat down with Soham Daga, AI researcher at supermemory, for a deep dive into how machines actually remember you. what we get into: - why every AI forgets you - why forgetting is the hard part, not remembering - memory that understands time - going past reranking with aggregation - buckets for agent memory - SMFS, cutting tokens by 60% A conversation on the future of context, full episode below!

Everyone wants AI with a bigger memory, we think that's the wrong goal.. We sat down with Soham Daga, AI researcher at supermemory, for a deep dive into how machines actually remember you. what we get into: - why every AI forgets you - why forgetting is the hard part, not remembering - memory that understands time - going past reranking with aggregation - buckets for agent memory - SMFS, cutting tokens by 60% A conversation on the future of context, full episode below!

supermemory

20,600 просмотров • 1 месяц назад

Introducing Memory 101 🧠 Every ai tool you use is a goldfish. brilliant for one chat, close the tab and it forgets you completely. memory is the most underbuilt layer in ai right now. so we're explaining how it actually works, one episode at a time. coming up: → rag vs memory, why most "ai memory" is just search → the 5 things real memory needs: facts, profiles, graphs, time, forgetting → why good memory has to forget things follow along!

Introducing Memory 101 🧠 Every ai tool you use is a goldfish. brilliant for one chat, close the tab and it forgets you completely. memory is the most underbuilt layer in ai right now. so we're explaining how it actually works, one episode at a time. coming up: → rag vs memory, why most "ai memory" is just search → the 5 things real memory needs: facts, profiles, graphs, time, forgetting → why good memory has to forget things follow along!

supermemory

17,235 просмотров • 3 месяцев назад

Episode 1 of memory 101: the anatomy of memory.🧠 real memory isn't one thing. it's five: > facts > profiles > relationships > temporality > forgetting "AI memory" products build one of these primitives and stop. Real memory is when all five compose. Next, stay tuned for RAG ↓

Episode 1 of memory 101: the anatomy of memory.🧠 real memory isn't one thing. it's five: > facts > profiles > relationships > temporality > forgetting "AI memory" products build one of these primitives and stop. Real memory is when all five compose. Next, stay tuned for RAG ↓

supermemory

23,127 просмотров • 3 месяцев назад

Demis Hassabis says bigger context windows are still a brute force answer to memory. The human brain does something stranger. During sleep, it replays what matters and folds new knowledge into what it already knows. AI does not need infinite context. It needs the right memory at the right moment.

Demis Hassabis says bigger context windows are still a brute force answer to memory. The human brain does something stranger. During sleep, it replays what matters and folds new knowledge into what it already knows. AI does not need infinite context. It needs the right memory at the right moment.

vitrupo

81,896 просмотров • 5 месяцев назад

Instinct is the AI assistant every VC is chasing. Its founder, Noah Shinn, co-wrote Reflexion, one of the earliest agentic frameworks with MIT's Ashwin Gopinath. Noah chose to build agents for individuals. Ashwin quietly went the other way and founded Sentra to bring memory to enterprise agents. Sentra is now the company brain of a $12B public company-- 18,000 employees, and the CXOs start their day on it. Nothing like it exists in public. I sat down with Ashwin at Land's End in San Francisco before any of this was revealed. We got into: •Why memory is not a knowledge base •Why organizational memory is easier to build than individual memory •Why he thinks the AI race is nowhere near over •Who's actually accountable when AI gets something wrong •What worries him about his daughter growing up unable to tell AI from a person 𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒 (00:00) Meet Ashwin at Land's End (00:48) What Sentra Builds (03:53) Why Agents Need Memory (06:44) Memory vs. Knowledge Base (08:38) Why Org Memory Works (13:02) Who's Accountable When AI Gets It Wrong (15:36) Humans Becoming AI (18:52) The AI Race Is Not Over (29:43) Choosing the Right Team (36:10) The AI Transition and Perspective (39:34) Kids Growing Up With AI (46:09) Agency as the Edge

Instinct is the AI assistant every VC is chasing. Its founder, Noah Shinn, co-wrote Reflexion, one of the earliest agentic frameworks with MIT's Ashwin Gopinath. Noah chose to build agents for individuals. Ashwin quietly went the other way and founded Sentra to bring memory to enterprise agents. Sentra is now the company brain of a $12B public company-- 18,000 employees, and the CXOs start their day on it. Nothing like it exists in public. I sat down with Ashwin at Land's End in San Francisco before any of this was revealed. We got into: •Why memory is not a knowledge base •Why organizational memory is easier to build than individual memory •Why he thinks the AI race is nowhere near over •Who's actually accountable when AI gets something wrong •What worries him about his daughter growing up unable to tell AI from a person 𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒 (00:00) Meet Ashwin at Land's End (00:48) What Sentra Builds (03:53) Why Agents Need Memory (06:44) Memory vs. Knowledge Base (08:38) Why Org Memory Works (13:02) Who's Accountable When AI Gets It Wrong (15:36) Humans Becoming AI (18:52) The AI Race Is Not Over (29:43) Choosing the Right Team (36:10) The AI Transition and Perspective (39:34) Kids Growing Up With AI (46:09) Agency as the Edge

Julia Fedorin

34,124 просмотров • 25 дней назад

AI has the IQ of a genius. And the memory of a goldfish. That’s why today’s AI still feels frustrating. But here’s the question nobody’s asking… What happens when AI remembers everything about you? And more importantly… Who owns that memory? Day 321.

AI has the IQ of a genius. And the memory of a goldfish. That’s why today’s AI still feels frustrating. But here’s the question nobody’s asking… What happens when AI remembers everything about you? And more importantly… Who owns that memory? Day 321.

Eric Jackson

36,201 просмотров • 3 месяцев назад

🤯 Your entire memory might NOT be inside your brain. We live in a Holographic Universe 😱 Neuropsychiatrist Diane Hennacy Powell of the just dropped this on Danny Jones Podcast “Memory is stored in a field — not in the brain. We’ve already proven it with rats.” Lashley cut huge parts of rats’ brains… and they still remembered mazes perfectly. Brain damage doesn’t erase memories the way we expect. What if your brain is just the TV — and the show is being broadcast from somewhere else? Holographic universe theory says every piece contains the whole. So… where are YOUR memories actually stored? ↓ Watch the clip — what do you think after seeing this? #Consciousness #Neuroscience #HolographicUniverse

🤯 Your entire memory might NOT be inside your brain. We live in a Holographic Universe 😱 Neuropsychiatrist Diane Hennacy Powell of the just dropped this on Danny Jones Podcast “Memory is stored in a field — not in the brain. We’ve already proven it with rats.” Lashley cut huge parts of rats’ brains… and they still remembered mazes perfectly. Brain damage doesn’t erase memories the way we expect. What if your brain is just the TV — and the show is being broadcast from somewhere else? Holographic universe theory says every piece contains the whole. So… where are YOUR memories actually stored? ↓ Watch the clip — what do you think after seeing this? #Consciousness #Neuroscience #HolographicUniverse

Interstellar

35,520 просмотров • 8 месяцев назад

Memory isn’t just a feature in AI, it’s the difference between a system that responds and one that truly understands. In the first episode of The Long Walk, Sudarshan Kamath and I dive into how memory is shaping the next generation of voice agents, and why it’s becoming critical for real enterprise workflows.

Memory isn’t just a feature in AI, it’s the difference between a system that responds and one that truly understands. In the first episode of The Long Walk, Sudarshan Kamath and I dive into how memory is shaping the next generation of voice agents, and why it’s becoming critical for real enterprise workflows.

Nishkarsh

30,316 просмотров • 7 месяцев назад

Most people use Claude Code like a chatbot. That is the mistake. The real power is not in better prompts. It is in controlling the session. /init gives the project memory. /memory keeps the rules alive. /clear removes old context. /compact saves the important parts. /context shows where the tokens go. /rewind fixes bad turns. /plan stops blind edits. /model matches the brain to the task. Claude Code gets 10x better when you stop letting the conversation run itself. Memory. Context. Boundaries. Rollback. Planning. That is the system. Stop asking Claude Code to “just do it”. Teach it what to remember, what to forget, and when to stop. Bookmark this before your next session turns into 40 messages of context debt.

Most people use Claude Code like a chatbot. That is the mistake. The real power is not in better prompts. It is in controlling the session. /init gives the project memory. /memory keeps the rules alive. /clear removes old context. /compact saves the important parts. /context shows where the tokens go. /rewind fixes bad turns. /plan stops blind edits. /model matches the brain to the task. Claude Code gets 10x better when you stop letting the conversation run itself. Memory. Context. Boundaries. Rollback. Planning. That is the system. Stop asking Claude Code to “just do it”. Teach it what to remember, what to forget, and when to stop. Bookmark this before your next session turns into 40 messages of context debt.

Fluixo

19,226 просмотров • 3 месяцев назад

The difference between memory and a second brain... In Hermes you have a memory.md file, but it's not particularly useful for storing a lot of info. Your second brain is a type of memory where you can distil down your raw inputs into insight you can use with AI. If you want to build a second brain with GPT6 Astra, check out the article I put together below.

The difference between memory and a second brain... In Hermes you have a memory.md file, but it's not particularly useful for storing a lot of info. Your second brain is a type of memory where you can distil down your raw inputs into insight you can use with AI. If you want to build a second brain with GPT6 Astra, check out the article I put together below.

Tom

24,164 просмотров • 24 дней назад

We launched our Company Brain!! 🧠 onboarding is a context problem: your first weeks you're not working, you're reconstructing what the team already knows. company brain is that context, queryable and able to act. shared memory for your team, in slack: → answers from the company's own knowledge, grounded and cited → acts in your tools as you, files the ticket and assigns it, with your approval → runs on a permissions graph, so it only reads with your access → chimes in only when it can help → same brain and permissions follow you into cursor and claude code when i joined supermemory, i braced for the onboarding tax. it never came. i just asked the company brain

We launched our Company Brain!! 🧠 onboarding is a context problem: your first weeks you're not working, you're reconstructing what the team already knows. company brain is that context, queryable and able to act. shared memory for your team, in slack: → answers from the company's own knowledge, grounded and cited → acts in your tools as you, files the ticket and assigns it, with your approval → runs on a permissions graph, so it only reads with your access → chimes in only when it can help → same brain and permissions follow you into cursor and claude code when i joined supermemory, i braced for the onboarding tax. it never came. i just asked the company brain

Muskan Jain

44,090 просмотров • 2 месяцев назад

In 2010, Stanford neurologist Frank Longo gave a masterclass on how learning & memory actually work. He explained: • The 7-item limit of short-term memory • Why sleep rewires your memories • How emotions make memories stick 20 lessons on how memory works(and why it fails):

In 2010, Stanford neurologist Frank Longo gave a masterclass on how learning & memory actually work. He explained: • The 7-item limit of short-term memory • Why sleep rewires your memories • How emotions make memories stick 20 lessons on how memory works(and why it fails):

Prasad

132,469 просмотров • 6 месяцев назад

"What happens when we can access each other's AI memory in enterprises?" scott belsky, investor/founder of Behance: "If I was working as an intern for years, and the context window is accessible to the enterprise, can the enterprise (keep prompting) and asking me what I'd think or do, after I leave?" "Every company right now is trying to build this AI memory for each of its products and users. A lot of them are building connectors to form a collective memory." "In the social/consumer world, imagine your girlfriend saying she wants access to your context window or memory." "There are all kinds of weird questions and implications that arise from this."

"What happens when we can access each other's AI memory in enterprises?" scott belsky, investor/founder of Behance: "If I was working as an intern for years, and the context window is accessible to the enterprise, can the enterprise (keep prompting) and asking me what I'd think or do, after I leave?" "Every company right now is trying to build this AI memory for each of its products and users. A lot of them are building connectors to form a collective memory." "In the social/consumer world, imagine your girlfriend saying she wants access to your context window or memory." "There are all kinds of weird questions and implications that arise from this."

TBPN

44,908 просмотров • 1 год назад

STANFORD BUILT AN LLM MEMORY THAT WORKS LIKE A BRAIN AND IT MADE STANDARD RAG LOOK ANCIENT no fine-tuning. no bigger model. just a smarter way to remember the insight: the brain doesn't store memories as isolated notes. it indexes them and links them - so one cue pulls back the whole memory so HippoRAG stops treating documents as separate chunks. it builds a knowledge graph, then spreads activation across it (Personalized PageRank) to pull a full answer from a single cue - the brain calls this "pattern completion" the result: > up to 20% better on hard multi-hop questions > 10-30x cheaper and 6-13x faster than iterative retrieval > connects facts split across documents in a single step most RAG still encodes every passage in isolation - so it breaks the moment the answer is scattered more retrieval was never the answer. brain-like memory is bookmark this. the next leap in AI agents is how they remember, not how big they get

STANFORD BUILT AN LLM MEMORY THAT WORKS LIKE A BRAIN AND IT MADE STANDARD RAG LOOK ANCIENT no fine-tuning. no bigger model. just a smarter way to remember the insight: the brain doesn't store memories as isolated notes. it indexes them and links them - so one cue pulls back the whole memory so HippoRAG stops treating documents as separate chunks. it builds a knowledge graph, then spreads activation across it (Personalized PageRank) to pull a full answer from a single cue - the brain calls this "pattern completion" the result: > up to 20% better on hard multi-hop questions > 10-30x cheaper and 6-13x faster than iterative retrieval > connects facts split across documents in a single step most RAG still encodes every passage in isolation - so it breaks the moment the answer is scattered more retrieval was never the answer. brain-like memory is bookmark this. the next leap in AI agents is how they remember, not how big they get

NO1ennn

32,454 просмотров • 1 месяц назад

What is MCP and why is everyone talking about it? In our latest episode of the GitHub Podcast, the team breaks down how the Model Context Protocol is making AI tools more powerful, open, and accessible. Listen wherever you catch your podcasts or

What is MCP and why is everyone talking about it? In our latest episode of the GitHub Podcast, the team breaks down how the Model Context Protocol is making AI tools more powerful, open, and accessible. Listen wherever you catch your podcasts or

GitHub

25,233 просмотров • 1 год назад

Yesterday, I joined TBPN to talk about what it really means to give AI a memory. And, why that’s becoming the next big moat. We discussed: - Why today’s AI apps are stateless - How memory is quickly becoming the real moat for AI systems - How Mem0 is decoupling memory from models - Why users should be able to own and carry their AI memories across apps - And, how 200 cold emails changed my life🙂 Appreciate John Coogan, Jordi Hays & the TBPN team for having me and for such a thoughtful discussion around the future of AI infrastructure.

Yesterday, I joined TBPN to talk about what it really means to give AI a memory. And, why that’s becoming the next big moat. We discussed: - Why today’s AI apps are stateless - How memory is quickly becoming the real moat for AI systems - How Mem0 is decoupling memory from models - Why users should be able to own and carry their AI memories across apps - And, how 200 cold emails changed my life🙂 Appreciate John Coogan, Jordi Hays & the TBPN team for having me and for such a thoughtful discussion around the future of AI infrastructure.

Taranjeet

14,913 просмотров • 11 месяцев назад

MOST PEOPLE USE CLAUDE CODE LIKE CHATGPT WITH A TERMINAL That is why they only get 30% of the value. They paste a task, wait for an answer, then keep explaining the same rules again when the context gets messy. The real power starts when you stop treating it like a chatbot and start treating it like an operating system for work. The useful commands are not flashy. They just remove the parts that quietly waste hours: > /init builds the project memory > /memory edits the rules Claude should keep > /clear resets the chat without losing the setup > /compact compresses a long session before quality drops > /context shows where your tokens are going > /rewind rolls back when an edit breaks the direction > /plan forces a blueprint before execution > /model matches the brain to the task > /goal lets it work toward a clear finish line This is the difference between asking AI for help and actually managing an AI worker. One mode gives you random useful answers. The other gives you memory, boundaries, checkpoints, context control and repeatable execution. Most people are still trying to write better prompts. The better move is building a better system around the model. save this

MOST PEOPLE USE CLAUDE CODE LIKE CHATGPT WITH A TERMINAL That is why they only get 30% of the value. They paste a task, wait for an answer, then keep explaining the same rules again when the context gets messy. The real power starts when you stop treating it like a chatbot and start treating it like an operating system for work. The useful commands are not flashy. They just remove the parts that quietly waste hours: > /init builds the project memory > /memory edits the rules Claude should keep > /clear resets the chat without losing the setup > /compact compresses a long session before quality drops > /context shows where your tokens are going > /rewind rolls back when an edit breaks the direction > /plan forces a blueprint before execution > /model matches the brain to the task > /goal lets it work toward a clear finish line This is the difference between asking AI for help and actually managing an AI worker. One mode gives you random useful answers. The other gives you memory, boundaries, checkpoints, context control and repeatable execution. Most people are still trying to write better prompts. The better move is building a better system around the model. save this

Asteri

15,347 просмотров • 3 месяцев назад

Engineer runs a Kimi K3 memory layer that costs $11 a month and remembers what a $500,000 vector database keeps losing. No embeddings. Four nodes and one rule about what's allowed to be forgotten. He published the whole schema. His version starts from the opposite idea. Memory is not a pile you search. It's a set of claims that expire unless something keeps paying to keep them. Four nodes. Every memory carries a clock someone has to reset: > WRITER - stores a fact with the reason it mattered, never raw text > DECAY - ages every memory down. Silence is deletion > RENEWER - only re-lifts a memory the model actually used again > GRAVE - holds what died, and why nobody reached for it Three nodes keep memory alive. One keeps the dead ones. Recall isn't storage here. It's rent a fact has to keep earning. That's the entire design. When everything is remembered forever, the useful and the stale retrieve identically. He replayed two months of agent context. 90,000 stored facts. 71,000 never retrieved once. The vector store returned all of them on similarity. Similarity graded closeness. Nobody graded whether the memory was ever right. Everyone else stuffs more into the context window and calls it memory. He built a layer that lets a fact die unless it keeps proving itself. The cost isn't storage. It's finding out how much of what your agent "knows" it has never once used. The article below is the full build - node prompts, the decay curve, the renewal rule. Save it. You'll want it open in the other tab.

Engineer runs a Kimi K3 memory layer that costs $11 a month and remembers what a $500,000 vector database keeps losing. No embeddings. Four nodes and one rule about what's allowed to be forgotten. He published the whole schema. His version starts from the opposite idea. Memory is not a pile you search. It's a set of claims that expire unless something keeps paying to keep them. Four nodes. Every memory carries a clock someone has to reset: > WRITER - stores a fact with the reason it mattered, never raw text > DECAY - ages every memory down. Silence is deletion > RENEWER - only re-lifts a memory the model actually used again > GRAVE - holds what died, and why nobody reached for it Three nodes keep memory alive. One keeps the dead ones. Recall isn't storage here. It's rent a fact has to keep earning. That's the entire design. When everything is remembered forever, the useful and the stale retrieve identically. He replayed two months of agent context. 90,000 stored facts. 71,000 never retrieved once. The vector store returned all of them on similarity. Similarity graded closeness. Nobody graded whether the memory was ever right. Everyone else stuffs more into the context window and calls it memory. He built a layer that lets a fact die unless it keeps proving itself. The cost isn't storage. It's finding out how much of what your agent "knows" it has never once used. The article below is the full build - node prompts, the decay curve, the renewal rule. Save it. You'll want it open in the other tab.

wast3

67,778 просмотров • 1 месяц назад

Happy to properly launch Anna, the proactive AI agent for parents! Uncovering a bit of the technology behind the scenes! Building Anna is where I learned: 💾 Memory as plain text sucks. You need structured memory. Like a full-blown PostgreSQL DB that stores your tasks and calendar in a structured manner. Most harnesses are good at coding-related stuff. Let it do the query. Don't let it vibe-search the memory. Let it vibe your SQL query 💭 Dreaming is a useful concept for enhancing memory to feed the LLM context. But DO NOT vibe your dream. Asking your agent to "hey, just dream and keep the relevant memory around" is a recipe for deleting a bunch of important information and keeping trash around. Your dream needs to have some Taxonomy (or better, Ontology). What information is important? For who? With what object? What can they do? And again, these are impossible to describe and act well without a proper schema 🔄 Loop Engineering is important for smoothing out rough edges in the system we build. But even expensive loop engineering with a state-of-the-art model can't out-engineer bad system design. The highest leverage an AI Engineer can do is actually building the right system design, and having an eye on both product delight and engineering scalability There are several more insights that I plan to cover in a dedicated video about Agentic AI Engineering. But it's actually a huge relief that the future of software engineering... is still software engineering

Happy to properly launch Anna, the proactive AI agent for parents! Uncovering a bit of the technology behind the scenes! Building Anna is where I learned: 💾 Memory as plain text sucks. You need structured memory. Like a full-blown PostgreSQL DB that stores your tasks and calendar in a structured manner. Most harnesses are good at coding-related stuff. Let it do the query. Don't let it vibe-search the memory. Let it vibe your SQL query 💭 Dreaming is a useful concept for enhancing memory to feed the LLM context. But DO NOT vibe your dream. Asking your agent to "hey, just dream and keep the relevant memory around" is a recipe for deleting a bunch of important information and keeping trash around. Your dream needs to have some Taxonomy (or better, Ontology). What information is important? For who? With what object? What can they do? And again, these are impossible to describe and act well without a proper schema 🔄 Loop Engineering is important for smoothing out rough edges in the system we build. But even expensive loop engineering with a state-of-the-art model can't out-engineer bad system design. The highest leverage an AI Engineer can do is actually building the right system design, and having an eye on both product delight and engineering scalability There are several more insights that I plan to cover in a dedicated video about Agentic AI Engineering. But it's actually a huge relief that the future of software engineering... is still software engineering

Gogo | Dota for Toxicity

30,766 просмотров • 3 месяцев назад

researchers gave a tiny local model human-style memory and its context limit basically stopped existing a team from MBZUAI, Princeton and Weizmann took a 1B model and rebuilt how it reads. instead of attending to everything at once, the model reads in 1,024 token chunks and passes the important stuff forward through an associative memory, the same way you carry the plot of a book between chapters without rereading them. the design mirrors human memory on purpose. full attention inside a chunk works as short-term memory. the module that carries information between chunks works as long-term memory. they even trained it like a person, starting with short easy texts and raising the difficulty gradually, because memory thrown into the deep end learns nothing. the numbers back it up. the normal model burns 40GB of GPU memory on a long document and collapses hard past its limit, dropping from 0.86 to 0.32 accuracy. the memory version holds 0.71 at double that length while using a flat 12GB no matter how long the input gets. it also needs about 30% fewer FLOPs. the part i keep thinking about is that nobody scaled anything here. they didn't build a bigger model, didn't stretch the window, didn't add compute. they looked at how a brain handles a long day and copied the architecture. a model small enough to run on a consumer gpu now survives documents its own architecture used to choke on. we keep treating intelligence as a compute problem. sometimes it's a memory problem.

researchers gave a tiny local model human-style memory and its context limit basically stopped existing a team from MBZUAI, Princeton and Weizmann took a 1B model and rebuilt how it reads. instead of attending to everything at once, the model reads in 1,024 token chunks and passes the important stuff forward through an associative memory, the same way you carry the plot of a book between chapters without rereading them. the design mirrors human memory on purpose. full attention inside a chunk works as short-term memory. the module that carries information between chunks works as long-term memory. they even trained it like a person, starting with short easy texts and raising the difficulty gradually, because memory thrown into the deep end learns nothing. the numbers back it up. the normal model burns 40GB of GPU memory on a long document and collapses hard past its limit, dropping from 0.86 to 0.32 accuracy. the memory version holds 0.71 at double that length while using a flat 12GB no matter how long the input gets. it also needs about 30% fewer FLOPs. the part i keep thinking about is that nobody scaled anything here. they didn't build a bigger model, didn't stretch the window, didn't add compute. they looked at how a brain handles a long day and copied the architecture. a model small enough to run on a consumer gpu now survives documents its own architecture used to choke on. we keep treating intelligence as a compute problem. sometimes it's a memory problem.

Alex Veremeyenko

16,147 просмотров • 2 месяцев назад