Jon Berry (@jonberry) header
Jon Berry (@jonberry) profile picture

BerrySmart / https://teapawt.net/t.co/zZ5KskDUU7 / https://teapawt.net/t.co/olLKaA2b9g / https://teapawt.net/t.co/EtOqsZIjlv / https://teapawt.net/t.co/TsjdKcwK5i - Strategic Problem Solver, Nostalgic & Hopeful

West Milford, NJ
Joined May 2007
Photo by Jon Berry (@jonberry)Photo by Jon Berry (@jonberry)Photo by Jon Berry (@jonberry)Photo by Jon Berry (@jonberry)Photo by Jon Berry (@jonberry)
Jon Berry retweeted
出大事了,兄弟们 !! DeepSeek 再一次给全世界人民送大礼,牛逼之举胜过 Anthropic 1000000倍 就在刚刚,DeepSeek 放出一个开源库,deepseek-ai/chat-demo 它可不是个套壳: 1/ 不需要写一行代码;不要买任何服务器;更不需要平凑第三方插件 2/ 原生边想边搜,直接调用官方内置 deepseek_web_search, 深度思考模式下,多步推理+全网检索+信源引用 3/ 全能文档/网页绞肉机:PDF,Word,PPT,图片拖进去秒解析,URL 自动抓文 4/ 原生语言交互:自带移动端实时音频输入 + 回答朗读 5/ 极致隐私保护和数据自由:所有数据存档本地 IndexedDB 什么是开源精神:最强模型权重 + 官方满配武器库,全部交到你手里,充分相信使用者对社区,对人类的贡献 爱折腾的人类,立即马上 Fork, 世界有你而精彩: github.com/deepseek-ai/chat-…
97
600
37
4,595
407,536
Jon Berry retweeted
Today we’re introducing Tofu-1, Bland’s speech-to-speech model. The highest-stakes calls rarely happen in good conditions: from moving cars and hospital rooms, over weak signal, with several people talking at once. Tofu-1 is built for those calls. It listens while it speaks, reasons while it talks, and can tell a caller apart from everything else on a line, like a TV or a crying baby. Join the preview: bland.ai/speech-to-speech?ut…
194
372
112
3,099
1,717,894
Jon Berry retweeted
Scientists just cracked the multiple sclerosis code after decades of searching. Two specific gut bacteria are triggering the disease, and they've proven it using identical twins and mice. This changes everything we know about the pathology.
53
524
22
2,661
125,485
Jon Berry retweeted
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
1,813
6,405
2,188
41,143
6,640,882
BREAKING: Russia delegation chief says accord reached with Ukraine on major prisoner exchange
3
37
6
234
21,951
Jon Berry retweeted
Web4 is here 😂
Web4 will be finally closing our computer and going outside
46
32
13
1,075
83,512
Jon Berry retweeted
Before movies had sound, playing background music was key for developing a particular mood for a certain scene. The photoplayer, an automatic mechanical orchestra, was frequently used by theaters [🎹 Joe Rinaudo, 🎶 California Gold]
39
154
21
803
73,188
Jon Berry retweeted
Brilliant solution for lighting fireworks safely at a distance
22
12
2
197
73,975
Jon Berry retweeted
Exa now indexes 3,000,000+ podcasts. Your agents can now search across millions of hours of transcripts, updated daily, and track what the world is talking about.
56
94
58
1,638
256,723
Jon Berry retweeted
Deepseek just revealed every AI agent secret Incredibly important article to read if you're building anything AI agent related Here's basically everything you need to know about building a Grok Bot quality AI agent: 1. Agents are split between routers and workers. The user talks to the router, then gives the work to the workers. The router never does work. This is why the chief of staff model in Grok Bot works SO well 2. Proactive means acting on events, not chats. The agent should always be watching for events and react based on every event that comes through it. Routing to the right worker/action 3. Messages should be in a queue. Users should be able to send messages even while the agent is working, and each message queues/steers the agent 4. Besides the router and workers, there should be a watchdog layer. The watchdog makes sure the agent is always up and fixes it when there are issues. This is why Hermes succeeded and OpenClaw struggled. Constantly broke and couldn't fix itself 5. Protect human attention. Only message the user when it needs to act or the user can use the result. Lighten their cognitive load. Another thing Grok Bot is excellent at. When you message it you can be confident the next message you get will be it saying the task is complete. Carve off an hour and read this incredible article
155
94
11
1,297
142,306
Jon Berry retweeted
Ten million people have watched a comedian accidentally destroy the entire TED talk industry. He filmed the talk once at TEDxNewYork in 2014 and openly admitted he had nothing to say. Speaking coaches charge $5,000 to teach the exact tricks he used in 6 minutes to say nothing at all. His name is Will Stephen. He is a comedian from Late Night with Seth Meyers who was invited to give a real TED talk and used it to prove the format is a costume anyone can wear. Every TED speaker opens with a question, tells a personal story, cites a scientist, and shows a chart. Stephen did all four in six minutes with zero content. His entire formula fits on a napkin. Gesture with both hands. Ask the audience a question anyone can answer. Tell a mildly embarrassing story so they say aww. Point to a stranger on a slide and call them important. Flash a chart of random numbers. Wear glasses with no lenses. The audience will think you are brilliant. That last trick alone has probably cost the public speaking coaching industry a hundred million dollars in sessions clients should have skipped. "If you watched me on YouTube with the sound off, you would think I was a genius." That is Will Stephen on the TED stage, three minutes into the talk he was paid to give. He wasn't lying. Executives pay $15,000 for a keynote coach and still open with a joke the room never laughs at. Founders rehearse their pitch for weeks and still sound like Stephen at minute four, saying "wagga wah, gabba gabba" with authority. The talk is free on YouTube. The method is written across every TED speaker's playbook for ten years. Stephen still writes for Seth Meyers. Almost none of the ten million viewers have stopped nodding at empty slides since. The formula is free. The willingness to actually say something before your next keynote, pitch, or townhall is the entire edge.
1,059
6,676
982
52,404
8,383,165
Introducing Reality-Aware Rendering in @Simulon, a completely new way to create with the real world. Place neural assets with your phone and act out the scene. When you stop recording, Simulon renders the take, reading everything in it (what you do, what you say and the place around you), and your assets respond to all of it. The final is regenerated from your recording: a new performance of the scene, not a copy of the footage. Available now on iPhone. Coming to Android soon.
938
2,576
946
15,656
1,701,827
Jon Berry retweeted
We're thrilled to announce the Superpower Peptide Library. Every peptide, mapped to the body. 152 peptides, from approved drugs to compounds still in trials, sorted by where they act. Each one has its own page, reviewed by a physician and fully cited, covering: - the evidence and how strong it is - its FDA status - it's mechanism of action - what studies in humans have found Because not all peptides are equal. Semaglutide is FDA-approved. BPC-157, maybe the most talked-about peptide for recovery and repair, is not. In a 2025 review of its research, 35 of 36 studies were in animals or the lab. Know the science before you buy the hype. Check it out here: superpower.com/guides/peptid…
215
838
157
8,703
1,462,268
We are developing a drug that opens up the capillaries and fills the body’s tissues with a compound that - according to our partner military labs - destroys all known pathogenic bacteria, all known pathogenic fungi, biofilms, influenza virus as well as viruses of the herpes family. These are facts. The FDA recently designed this product a Qualified Infectious Disease Product (QIDP) for all cause pneumonia, giving us FDA Fast Track, and Priority Review. Good news for the day. ❤️‍🩹🫡
MICROSCOPY. Making the invisible visible since 1590. (Yes, those little white balls are actually immune cells flowing through the tinniest blood vessels! Imaged painlessly, non-invasively through the skin)
173
501
64
4,528
395,983
Jon Berry retweeted
Yesterday I spent all day typing this into Claude Code: "Do a full security audit of this website and server" On every single VPS (and site) I have I already had automatic unattended upgrades with auto reboot for security updates in Ubuntu (reboot only takes 4 minutes so I think it's fine every few weeks) Also isolated each VPS on Tailscale with special security policies (generated by Claude Code! so it can not access any other VPS (in case it gets hacked) and not even access my MacBook or iPhone unless I talk to it first Also lots of user permission stuff isolating stuff to www-data to avoid root access for web services Not the most fun stuff but essential I think in addition to @yongfook's "marketing week" we now have to add one week per month of "security week" because with the models getting better our pre-emptive defense has to get better too Luckily for us I think the attack and defense capabilities of AI models seems to somewhat match? So as long as we keep typing "do a security audit" regularly into our clanker we can keep it safe I hope 🫨
The last two weeks have been probably the most stressful weeks of my entire adult life. ...while I was on "holidays" for the first time I can remember. Some parts were absolutely fucked: - I had to visit the ER in the US (fml) because I suspected having a stroke. I'm luckily fine, so no need to worry. It turns out the next point below might play a role in this - One of my companies has been AI cyber attacked unlike anything I've ever seen before. I got everything under control but given the attack originated from an AI and not a human it caused me so much stress to patch/fix on the spot that I still feel feel the aftermath physically today. This happened already a week ago so all the Black October chaos didn't surprise me at all. I literally saw it coming. Scary times folks but I'm sure we're gonna make it. Some parts were absolutely great: - Been spending a lot of quality time with my GF, exploring Taiwan and the US. - Lots of bucket list things got checked. Got a chance to see a show at the OG Comedy Store, somehow saw Ali G live (lol) and a few more. - Got to meet bunch of cool friends including @tommoor and @csallen - Finally bought a new MBP and some other stuff that's hard to get outside the US - I rediscovered my love for analog film. It's always been there but it was quite dormant past few years. I have a lot of cool ideas I want to pursue soon. - Lastly, some really great news I can share soon which also might have contributed to the insane stress the past weeks because I had to work until the early morning hours... Aight, that's my short personal life update. Hope ya'll happy, healthy and safe out there friends.
133
40
66
1,393
179,170
Jon Berry retweeted
This is sick
You can dogfight in my realistic fighter jet simulator now. viperoneone.com
69
46
7
1,744
164,347
WOW did we really just get a Mythos-level open-weight model?? in OCTOBER?! 🤯 oh—hi there, past pliny! spot-on predictin, well struck fren! 🍻 and right on time 🎃👻 dunno how you do it! *takes a deep breath* everything IS going to be fine, isn’t it? thanks for the reminder 🤗 onwards! 🫡 ⊰-•-•✧•-•-⦑/L\O/V\E/,\P/L\I/N\Y/⦒-•-•✧•-•-⊱
Replying to @elder_plinius
if i were a gambler, i'd put my money on Halloween. can't wait to quote tweet this. hi frens from the future! take a deep breath! everything is going to be fine! ⊰•-•✧•-•-⦑/L\O/V\E/,\P/L\I/N\Y/⦒-•-•✧•-•⊱
86
143
21
2,608
103,888
Jon Berry retweeted
DO YOU BELIEVE LUCIFER CONTROLS THE MUSIC YOU HEAR TODAY?
396
3,043
142
12,308
297,561
Jon Berry retweeted
What happens when Lethal Weapon meets Step Brothers.
7
104
4
883
52,082