The author used Grok Bot to generate the MemT-1.0 popular-science animation video purely with code

Holy crap bros, this is seriously awesome—it feels like the most frustrating seven-year amnesia history since the birth of large models may finally be turning the page!! I'd even go so far as to say something outrageous: this is probably the sexiest, yet most industry-critical, underlying breakthrough in large-model architecture evolution this year. MemT-1.0, just open-sourced, for the first time welds a true biological-grade hippocampus into the bottom layer of the Transformer, letting AI truly evolve from a goldfish that forgets the moment a chat ends into a life form that can precipitate experience into its very blood. That Paleolithic era of brute-forcing GPUs and propping up long context with several GB of KV Cache was officially smashed through today by this open-source native-memory large model. Their 66-page technical report and open-source weights of up to 288B directly throw out a set of dimensional-strike engineering bills: after turning memory into a native state of the forward computation flow, the VRAM overhead under a 256K ultra-long history plunges from 5.06 GB to a constant 40 MB, cutting the KV cache by more than a full 120 times, and the first-token latency stays steadily at just over 100 milliseconds no matter how long the history is stretched. What first breaks through the traditional path is its complete reconstruction of the model's forward computation flow. Previously, to make AI remember context, everyone could only grit their teeth and pile on long text or bolt on an external vector database, at the cost of blowing up VRAM and losing the causality of multi-step reasoning. It runs a native memory module in parallel beside the standard network; the moment input enters, the erase gate, write gate, and forget gate operate automatically like a biological hippocampus, no longer retaining all the dead text but digesting information in real time into a dynamic internal memory matrix. Even more stunning is the dimensional strike it brings in real agent scenarios: the team directly wrote over a thousand code experiences accumulated from historical error correction into the internal native memory state, occupying zero token budget of the system prompt, and as a result, on the standard code test set, the hardcore metric of passing three consecutive times directly improved by 8%, with stability even surpassing the approach of directly stuffing in plaintext prompts. This means experience is no longer a rigid rule the model must re-read and re-understand every time, but has been directly internalized into its instinctive intuition—truly awesome! I think this is the most fundamental physical logic on the path to true recursive self-improvement: parameters only define the model's factory lower bound, but only persistently internalized native memory can open up room for compounding growth in intelligence. Previously the models we bought were one-off computation tools that burn out after a chat; from today on, models are becoming self-evolving assets that understand the business better the more they are used and become more valuable the more they are used. When building agents or long-conversation applications, what gives everyone the biggest headache: the expensive VRAM and stuttering latency brought by long context, or the model's amnesia that forgets business rules the moment it turns its head? For me, just the single item of compressing 256K VRAM by 120 times with constant latency is already enough to instantly revive countless long-range businesses that previously couldn't make the numbers work hh~

View original

卧槽兄弟们，这个真的牛逼，感觉大模型诞生至今最憋屈的七年失忆史可能要翻篇了！！ 甚至我想说个暴论，这大概是今年大模型架构演进里最性感、但也最击中行业要害的一次底层突围。 刚刚开源的 MemT-1.0 第一次在 Transformer 底层焊入了一颗真正的生物级海马体， 让 AI 从聊完即忘的失忆金鱼，真正进化成了能把经验沉淀进血液里的生命体。 靠暴力堆显卡、拿几个 G 的 KV Cache 硬撑长文本的旧石器时代，今天被这个开源原生记忆大模型正式砸穿了。 他们放出的 66 页技术报告和最高 288B 的开源权重直接甩出了一组降维打击的工程账单： 把记忆变成前向计算流的原生状态后，256K 超长历史下的显存开销从 5.06 个 G 暴降到恒定的 40 兆， 直接把 KV 缓存砍掉了整整一百二十倍以上，而且首字延迟更是不管历史拉得多长，始终稳稳压在一百毫秒出头。 最先击穿传统路径的，是它彻底重构了模型的前向计算流。 以前大家为了让 AI 记住上下文，只能硬着头皮堆长文本或者外挂向量库，代价就是显存撑爆且丢失多步推理因果，它在标准网络旁并联了一套原生记忆模块，在输入进入的瞬间，擦除门、写入门和遗忘门就像生物海马体一样自动运转，不再保留所有的死文字，而是把信息实时消化成动态的内部记忆矩阵。 更绝的是它在真实智能体场景下带来的降维打击，团队把历史纠错中沉淀的一千多条代码经验，直接写入内部的原生记忆状态，完全不占用任何系统提示词的 Token 额度， 结果在标准代码测试集上，连续三次均通过的硬核指标直接提升了百分之八，稳定性甚至反超了直接塞明文提示词的做法。 这意味着经验不再是一段需要模型每次重新阅读理解的生硬规则，而是直接内化成了它的本能直觉，真是牛逼! 我觉得这才是通往真正的递归自我改进最底层的物理逻辑啊, 参数只定义了模型的出厂下限，但唯有持久内化的原生记忆，才能为智能打开复利增长的空间。 以前我们买的模型是聊完即焚的一次性计算工具，从今天起模型正在变成越用越懂业务、越用越值钱的自进化资产。 大家在做智能体或者长对话应用时，最头疼的是长上下文带来的昂贵显存和卡顿延迟，还是模型转头就忘业务规范的失忆症？ 对我来说，光是把 256K 显存压缩一百二十倍且延迟恒定这一项，就已经能让无数原本算不过账的长程业务直接原地复活了hh~