Dr Eam Co Ding

56.4K posts

Dr Eam Co Ding banner
Dr Eam Co Ding

Dr Eam Co Ding

@xDreamCoding

#LaserRayTil1k $TSLA zertifizierter #OstmullenDienstag Fan 🇩🇪

Katılım Haziran 2012
1.3K Takip Edilen1.4K Takipçiler
naiive
naiive@naiivememe·
Calming down bro after fomo-buying into Silver at $130 and now it's $66 , -55%
English
36
118
1.7K
101.9K
can
can@marmaduke091·
🚨 100M TOKEN CONTEXT WITHOUT COLLAPSE > <9% degradation from 16K → 100M > beats RAG + rerank + SOTA pipelines > runs on just 2×A800 GPUs we could be back
can tweet media
艾略特@elliotchen100

论文来了。名字叫 MSA,Memory Sparse Attention。 一句话说清楚它是什么: 让大模型原生拥有超长记忆。不是外挂检索,不是暴力扩窗口,而是把「记忆」直接长进了注意力机制里,端到端训练。 过去的方案为什么不行? RAG 的本质是「开卷考试」。模型自己不记东西,全靠现场翻笔记。翻得准不准要看检索质量,翻得快不快要看数据量。一旦信息分散在几十份文档里、需要跨文档推理,就抓瞎了。 线性注意力和 KV 缓存的本质是「压缩记忆」。记是记了,但越压越糊,长了就丢。 MSA 的思路完全不同: → 不压缩,不外挂,而是让模型学会「挑重点看」 核心是一种可扩展的稀疏注意力架构,复杂度是线性的。记忆量翻 10 倍,计算成本不会指数爆炸。 → 模型知道「这段记忆来自哪、什么时候的」 用了一种叫 document-wise RoPE 的位置编码,让模型天然理解文档边界和时间顺序。 → 碎片化的信息也能串起来推理 Memory Interleaving 机制,让模型能在散落各处的记忆片段之间做多跳推理。不是只找到一条相关记录,而是把线索串成链。 结果呢? · 从 16K 扩到 1 亿 token,精度衰减不到 9% · 4B 参数的 MSA 模型,在长上下文 benchmark 上打赢 235B 级别的顶级 RAG 系统 · 2 张 A800 就能跑 1 亿 token 推理。这不是实验室专属,这是创业公司买得起的成本。 说白了,以前的大模型是一个极度聪明但只有金鱼记忆的天才。MSA 想做的事情是,让它真正「记住」。 我们放 github 上了,算法的同学不容易,可以点颗星星支持一下。🌟👀🙏 github.com/EverMind-AI/MSA

English
28
96
1.4K
143.9K
MoreGainzs
MoreGainzs@moregainzs·
@Tesla I believe Rivian does that too 😆. This is not 2020 anymore.
English
20
3
40
4.8K
Tesla
Tesla@Tesla·
We – design the chips & hardware – make the cars w/ said hardware – collect real-world data at scale – train the real-world AI model – built (& continue to expand) the massive supercomputer cluster that trains it – deploy AI directly to millions of robots on wheels All that is shared with @Tesla_Optimus for broader applications in both the physical & digital world
English
558
1.5K
10.1K
696K
Dr Eam Co Ding retweetledi
Hodler
Hodler@TSLAshareholder·
you ask a $tsla bull why you should buy the dip and they dont even give you a reason anymore, they just look at you like this
Hodler tweet media
English
32
18
300
9.3K
Sawyer Merritt
Sawyer Merritt@SawyerMerritt·
Carwow just released a new video to its 11 million YouTube subscribers titled: "Why Tesla Full Self Drive is Pointless!" @carwowuk misleads its viewers into thinking Tesla’s Autopilot is FSD, even though FSD hasn’t been approved in the UK yet. Autopilot isn’t meant for city driving, yet they test of bunch of scenarios that Autopilot wasn't built to do in the first place....
Sawyer Merritt tweet media
English
507
265
3.8K
354.9K
Shaun Maguire
Shaun Maguire@shaunmmaguire·
Tesla FSD had its "Claude Code Moment" at the exact same time (roughly last Nov - Jan) Somehow folks are sleeping on the former, while being aware of the latter My wife and I bought two Teslas over the last quarter because of FSD IT'S INSANE 🚀 (...Next up, the Grok Moment?)
Elon Musk@elonmusk

Try Tesla FSD (self-driving)!

English
58
90
1.5K
112.1K
Dr Eam Co Ding retweetledi
Sebastian G
Sebastian G@SebastianG_De·
Sebastian G tweet media
ZXX
4
148
2.2K
18.2K
Dr Eam Co Ding
Dr Eam Co Ding@xDreamCoding·
@FrankfurtZack Alle "Rückrufe" sind einfache Over the Air updates. Etwas das die deutschen bis heute nicht hinbekommen ;)
Deutsch
0
0
1
48
Dr Eam Co Ding retweetledi
𝙲𝚘𝚠 𝙶𝚒𝚛𝚕 𝙺𝚒𝚜𝚜𝚎𝚛
European elections be like: The No Immigrants party got 49% of the vote narrowly losing to a coalition of Greens, Communists, Socialists, Democrats,and Center Right Party.
English
131
1.8K
38.8K
848.1K
Jo Bhakdi
Jo Bhakdi@JOBhakdi·
Tough times for the market when the government is becoming unhinged. But it's important to keep a cool head: AGI is here. $TSLA, $NVDA, Mag7, $MSTR, $HOOD and other AGI stocks are positioned to explode. The Iran disaster might look bad and mark the end of American empire - but at the same time we have to ask ourselves: does this really matter? The answer is: no. AGI companies are taking over the economy and will also increasingly determine policy and geopolitical order. I am not saying this as an endorsement of a total corporate takeover, but as a simple prediction and statement of fact.
English
15
1
58
4.5K
Jo Bhakdi
Jo Bhakdi@JOBhakdi·
I love how everyone (including me) who was leaning pro-Trump got sucked into wild theories at the beginning of the war what the "plan" was. My conclusion was simple: TACO is the plan to get some leverage with China. Why did I think that? Because it was the ONLY possible explanation that made sense. All other explanations make no sense, aka would mean Trump totally lost it. Turns out... he totally lost it. lol. From now, the only hope is that the overwhelming pressure and rapid deterioration of political capital makes him TACO anyway. But whoever makes such an enormous mistake once could also easily keep making that mistake. This is hard to predict at this point. The only thing we know is that China, Russia, Iran and Israel - for different reasons - have a vested interested in this not ending anytime soon. For Europe, this is clearly also a disaster - but there is no doubt that European leaders will gleefully watch the US self-destruct. Their rational minds may tell them that's not that great for Europe, but the emotional dynamics are obvious.
English
35
5
62
6.3K
Robin Ebers | AI Coach for Founders
people think AI is going to keep getting cheaper it's not → GPT 5.4 costs more than 5.3-Codex → fast mode is 2x on top of that → juicy ChatGPT 2x plan limits are over soon and claude limits are laughable to begin with the frontier gets more expensive from here even budget models like MiniMax get more expensive if you won't build it today... ... you might not be able to afford it in the future
English
53
1
74
5.7K
ZUBY:
ZUBY:@ZubyMusic·
It feels like the overall experience of social media has dropped significantly in the last few months. It's not unique to this platform, but all of the ones I use. Am I alone in this sentiment?
English
699
72
2.6K
187.4K
Dr Eam Co Ding
Dr Eam Co Ding@xDreamCoding·
@AlfakevinE i think them taking their time for 14.3 is a good thing honestly. making sure there is not too much regression with thinking
English
0
0
1
22
Gluteus Maximus 🇹🇯🇰🇬🇺🇿🇰🇿🍣🥩🏋️
@xDreamCoding Possibly: A) 4.5x shipped for 14.1 (a big jump from 13.x) B) or 10x shipped for 14.1 ( a bit less likely) Curiously, 14.2 doesn’t seem to be a major improvement from 14.1 in terms of “miles to critical DE”. Imho, there is a chance 2x the parameters vs 14.2 ships in 14.3
English
1
0
0
18
kache
kache@yacineMTB·
it is laughable how bad gemini is
kache tweet media
English
67
8
516
56.8K
fabi
fabi@fabinulleins·
AMD makes so much right and the market is sleeping. I think we will see a massiv run really soon.
English
4
0
6
607