Gary Basin

55K posts

Gary Basin banner
Gary Basin

Gary Basin

@garybasin

currently generating 2,663 tok/sec

www Katılım Ocak 2014
4.2K Takip Edilen13K Takipçiler
Sabitlenmiş Tweet
Gary Basin
Gary Basin@garybasin·
Introducing agentboard. A fast web wrapper around tmux optimized to multiplex AI agent TUIs, w/ special support for iOS safari and mac w/ keyboard shortcuts Fun little weekend project as I've gotten sick of using tmux through Blink on my phone to get to my Claude and codex sessions on my Mac github.com/gbasin/agentbo…
Gary Basin tweet mediaGary Basin tweet mediaGary Basin tweet media
English
34
11
280
53.6K
Melian Refugee
Melian Refugee@escapefrommelos·
these heritable markings were most certainly visible to early humans, showing tribe, status, etc. as humans lost the ability to display and see these marks, we began to artificially paint the human body with decoration... to compensate for this Lack, all art was created
Melian Refugee tweet mediaMelian Refugee tweet mediaMelian Refugee tweet media
Crémieux@cremieuxrecueil

Everyone has stripes known as Blaschko's lines. These are normally not visible and are generally only present if there's an issue, chimerism, etc. But some birds can see them!

English
15
22
603
55.8K
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
If Israel and the US disengage soon*, I expect they'll return to war in weeks-months, again. Iran is severely degraded but it'll rebuild stockpiles over years. Can't have that. *the plan seems to be blowing up tunnels and gacha-rolling assassinations until the Strait is reopened
English
13
2
59
5.2K
Christian Catalini
Christian Catalini@ccatalini·
5/ Economist @JamesBessen reconstructed this history. Once the power loom was in place, 62% of subsequent productivity gains came not from better machines, but from better-skilled humans who could monitor more of them. “They were monitoring.”
Christian Catalini tweet media
English
2
1
9
757
Christian Catalini
Christian Catalini@ccatalini·
1/ This is a great description of what verification infrastructure looks like in practice. In our new paper we argue this is the binding constraint on the AI economy — the same bottleneck textile mills hit when they scaled looms faster than weavers could check them.
Rohit@rohit4verse

x.com/i/article/2028…

English
8
24
197
51.5K
Gary Basin
Gary Basin@garybasin·
@jxmnop Can’t you just put codex in a loop these days
English
0
0
31
3.3K
dr. jack morris
dr. jack morris@jxmnop·
Learning to write kernels might be the highest-ROI activity for displaced SWEs: → prereq: reasonable engineering ablity → six to twelve months of study → millions of dollars, mark zuckerberg showing up at your house to hire you, etc. i wish this were an exaggeration
English
29
37
1.3K
72.2K
Roko 🐉
Roko 🐉@RokoMijic·
@garybasin I don't think lasers are very good against fast drones. Countermeasures such as mirror armor and short dwell time will render lasers ineffective.
English
6
1
55
6.3K
Roko 🐉
Roko 🐉@RokoMijic·
Imagine a swarm of 100 of these, all with explosive payloads, attacking you from all directions and elevations at 400 km/hr. That is the ultimate drone defense challenge and I think you need some kind of rapid fire AI controlled gun battery to succeed. You probably have 3 seconds to shoot in total.
Samuel Cardillo@CardilloSamuel

direct kinetic impact. a flying sword. 450km/h. updated video showing exactly that. we're also working on the explosive variant. only for authorized partners. dms are open.

English
49
19
344
37.6K
sucks
sucks@powerbottomdad1·
babies coming in the morning. please say a prayer for my wife and family. or send me 1000 USD. whatever your heart tells you
English
34
0
657
8.3K
Vladimir
Vladimir@vlelyavin·
@yacineMTB this is one of those predictions where the technical path is plausible but the licensing clusterfuck would be legendary
English
2
0
3
421
kache
kache@yacineMTB·
prediction: someone is going to get a coding AI like codex to automate turning existing steam video games into harnesses, come up with architecture to parallelize the games themselves in a manner that is conducive for RL training, and train an RL demigod model
English
35
12
412
18.6K
forward deployed ccp gf
forward deployed ccp gf@FangYi11101·
There are tech workers making $500k+ deciding offers based on whether they can take home leftover snacks from the office pantry
forward deployed ccp gf tweet media
English
60
21
2.3K
155.6K
Jake
Jake@nayshins·
@garybasin I'm trying to crowdsource some slop ok
English
2
0
9
553
Jake
Jake@nayshins·
Has anyone documented all the code slop patterns yet? I want to lint for them and banish them to hades.
English
39
3
200
21.8K
Gary Basin
Gary Basin@garybasin·
If they’re blowing up all the oil why isn’t the price going up? There must be something in the Dune book about this
English
3
1
28
1.7K
Gary Basin
Gary Basin@garybasin·
@inductionheads This reads like a hack that will only help with certain benchmarks though
English
1
0
2
451
Super Dario
Super Dario@inductionheads·
1) what
Super Dario 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
9
22
449
52.9K
Gary Basin
Gary Basin@garybasin·
@banteg It’s beautiful. What antidote will they be selling us
English
0
0
1
220
NIK
NIK@ns123abc·
🚨FBI: “Day One Ventures @mashadrokova was in Silicon Valley to steal technology…”
NIK tweet media
English
39
60
830
296.1K
Speculator
Speculator@TheSpeculator0·
Being on gardening leave in europe is eye opening. Go outside on a moderately nice wednesday and parks and terraces are just beaming with working age people. cool
English
1
2
201
16.7K