Tweet épinglé
Keeping the receipts.
Brad@BraddrofliT
Bookmark this: November 6, 2024 Stock Market 43,492 Annual Inflation 2.2% Unemployment 4.1%
English
hp
9.6K posts
@DeepValueBagger Will sub. Don’t have time to check the market this year with a lot of things going on with my life.
English
I'm really tired of arguing with random people in my posts. I have been thinking about creating a smaller circle, "DVB Circle". I would like to share more long form format. I have a couple thesis I think could be big. I have a really nice research pipeline to produce great reports to share to.
But i hate to be that guy with a sub. And I hate to be the guy having to nuture it. You guys know I don't need the money but it would be nice if I have a fund to increase my local LLM hardware.
The idea is
1) I give you excellent reports from my research pipeline.
2) private discord for like-minded people
3) small fee to subsidize my local LLM hardware + cloud fees etc.
4. You get some really interesting thesis on my mind.
English
@DeepValueBagger I am more bullish on SpaceX now. (I don’t own any space stocks.) People underestimate how hard it is for space business. I see many space stocks would be wiped out like Virgin Galactic in the next few years.
English
One of the moment I got bearish on $asts when i found out that they were going to launch with India space program. I dont know what happened to it but you'd think blue origin would do better. These are the risks... It's a matter of time.
Daniel Koss@daniel_koss
🚨Official news directly from $ASTS ASTS loses BlueBird 7 after New Glenn 3 mission placed satellite into a lower than planned orbit. - Satellite separated and powered on but altitude is too low for on-board thrusters to sustain operations; will be de-orbited - Cost expected to be recovered under the company's insurance policy - BlueBird 8-10 expected ready to ship in approximately 30 days; company currently in production through BlueBird 32 - $ASTS maintains target of approximately 45 satellites in orbit by end of 2026, with an orbital launch expected every 1-2 months on average
English
所以这解决了我心中久久挥之不去的问题：训练大语言模型真的很难吗？
应该能是不难的，门槛应该是很低的
否则解释不了为什么中国能有几百个大语言AI创业公司
关键的问题说到底，应该还是数据
然后现在硬件设施的皮瓶颈变成了推理，Agent，内存和CPU
Berryxia.AI@berryxia
Karpathy重磅观点：前沿AI模型之所以巨大，不是技术太复杂，而是训练数据全是垃圾！ 1. 真实预训练数据集里全是股票代码、坏HTML、spam和乱码，根本不是高质量内容 2. Llama 3 信息压缩率仅0.07 bits/token，大部分参数都在做“记忆”而非真正认知 3. 建议彻底分离：只保留高质量数据训练的认知核心（1B参数即可达真正智能）+ 可查询的外部记忆 4. 趋势已验证：GPT-4o 200B参数已全面超越原1.8T GPT-4，推理成本280倍下降 数据质量才是当前AI最大瓶颈！
中文
@jamiequint @vcjesse1 You are underestimating the execution moat.
I was on a Handshake project for two weeks and they still could not fix the tasking system. On the other hand, Mercor engineers can fix the tasking system on the fly when I complained to the TL an hour ago.
English
@vcjesse1 The data doesn't have to be easy to acquire for this to be true, it just has to be close to as difficult for everyone. Being slightly better at recruiting subject matter experts for gig labor is not a moat
English
On Mercor, Handshake, Surge AI, Turing, Micro1, etc.
If you're considering making a career move I would not go work at any of these companies.
These businesses are a race to the bottom. There is no structural defensibility in outsourced research/training data for a set of 5-10 very large customers.
Once the frontier labs have to start caring about margins, one of two things will happen: they will drive vendor margins down until these businesses trade at 0.5–1x revenue, like every other staffing company (potentially worse because of the customer concentration risk), or they vertically integrate.
All arguments about how these are important outsourced research companies that the labs really value are cope.
The Information@theinformation
Exclusive: Handshake’s gross annualized revenue from AI training has risen to nearly $1 billion, up from $550 million in January, a few months after it started the new business. Read more: thein.fo/3Ot5fdI
English
@darrencao2024 微软说到底saas占很大部分，这些office以后还有没有用肯定是个问题，如果以后办公都是agent，还要office干什么？直接交换mark down不是更简单？
中文
微软的资金量，有没有可能被anthropic吞掉？还是打不过就加入，甚至是打不过就买断？之前跌到370的时候，市场情绪就是微软老了，会被取代，现在这种声音怎么没了？微软目前的价位，仍然具有不错的潜力，看好年底前回到500以上。
价值&投资@joely7758521
$MSFT 开始用 Claude Opus 4.7 Open是好事🤔
中文
这样的帖子会误导很多投资者，就好像前几周下跌，类似的帖子满天飞，很多经验不足的投资者看到这样的帖子，第一时间肯定割肉离场。你靠发这样的帖子博流量，心里有愧疚吗
Kai@Kai866
2000年4月17日纳斯达克指数创下有史以来最大的单日涨幅，大家都在狂欢互联网科技的极大繁荣。 之后美股指数级别下跌了78%，直到30个月以后才完成真正的触底，更不要说当年的龙头思科在底部徘徊了近15年。 2026年4月17日有什么不一样吗？
中文
who is actually winning the AI race
OpenAI — $25B revenue but still not profitable
Anthropic — best model, just scared every government on earth
Google — most resources, losing to everyone on vibes
Meta — spending $130B and still playing catchup
xAI — Elon promised to match everyone by 2026
be honest
who wins
English
hp retweeté
This Allbirds story is so insane:
→ $BIRD IPO'd in 2021 at a $4 billion valuation
→ Silicon Valley's favorite shoe
→ Lost 99.5% of its value in 4 years
→ Closed every US store
→ Sold the entire brand for $39 million
→ Renamed itself "NewBird AI"
→ Using $50M to buy GPUs and compete with AWS
→ Stock up 450% today on 875x normal volume
This is the most unhinged corporate pivot of the decade and the newest meme stock entrant
Tracy Alloway@tracyalloway
Allbirds, the shoe brand, now says it's an AI compute company.
English
Anthropic just built their own Codex - the Claude way
Claude@claudeai
We've redesigned Claude Code on desktop. You can now run multiple Claude sessions side by side from one window, with a new sidebar to manage them all.
English
Dario is committing security fraud by presenting false statements about their model capabilities. @SECGov.
Chamath Palihapitiya@chamath
Creating fear porn and crying wolf does not serve the AI industry well. Maybe Anthropic actually believed what they were selling, in which case the naivety, if you can charitably call it that, is astounding. At a minimum, they need a few circumspect adults in the room that are detached enough from the financial outcome to help make these decisions in the future.
English
Anthropic's CEO: "coding is going away first. then all of software engineering.
So many guys have said this in this AI craze.
This ceo guy is pumping his AI, what should he say ?
thats why these guys should be taken with a tuckload of salt each time you listen to them
English
Anthropic will say whatever it takes to stay in the headlines.
Claude Mythos claims “thousands of vulnerabilities”
off just ~198 reviewed cases
• only ~198 reports actually reviewed
• “thousands” comes from extrapolation
• some bugs weren’t even practical to exploit
• some were already patched
• others blocked by basic defense layers
it's all marketing gimmick now.
English
One big problem with agentic coding today is that models are pretty “spiky.”
For example, Claude Opus is better at frontend + agentic workflows, while GPT-5.4 is better at backend + distributed systems.
But Claude Code and Codex are locked into their own models.
You also often have to jump between them. I sometimes write code with Claude, then when it has a complex bug, I have to spin up a separate terminal to have Codex review it. Ideally, you’d want multiple models collaborating within the same context.
Automatic model routing and cross-model or agent collaboration will be a huge unlock. There are a few technical challenges. Early model routers (like ChatGPT’s) were pretty rough.
(Cursor and OpenCode seem to be in the best position to do this. Let me know if they already have strong model routers.)
English