SV_Techie

@sureshsankaran @mukund @AnthropicAI yup. def miss on 1st wave. Now then can take Kimi and make it better 🤣

@sv_techie @mukund @AnthropicAI Wonder what happened to all the capex in $b spent so far..they dont even have good inhouse model and they are spending more dollars for claude sub..lol what a mess

$META now spends over $7B on @AnthropicAI Talk about customer concentration.

@SixSigmaCapital you just explained fintwits in 2 lines...🤣

If we rally watch all the people who haven’t shared their YTD since October start sharing screenshots again…

@mukund @AnthropicAI def they are spending a lot. They also track them really well.

@sv_techie @AnthropicAI $7 Billion is the estimate from my friend who works there from all the token maxxing

@mukund @AnthropicAI Probably not what is released to external world -- that is the reason they are heavily hiring OpenAI folks. Because they know spending $Bs on claude, they would rather train and use for internal use as well as to power IG. At least that is their current plan.

@sv_techie @AnthropicAI With Llama? What. I tried it on LMstudio it’s not great.

@KreizJordy Been saying Anthropic will become top spend by enterprise

The AI naysayers & luddites: ‘AI hasn’t proven its ability to contribute to revenue’ 🫧 Me: 😏

OpenAI investment returns: alienart.io/projects/69d33…

Cursor was so helpful in upgrading my old Ubuntu 22.* to 24 today. I ran into several issues due to various 3rd party pkgs I had -- if I had gone through good old method, there is no way I would have completed. Every step along the way when I run into issue cursor guided me.

@YingjunWu I don't hear much about Notion. Is it mostly used by marketing/product mgmt folks?

It really feels like the Notion team is doing AI just for the sake of doing AI, and they are trying to monetize it way too early by charging extra for all these AI features. What makes it worse is the pricing is based on the whole team. So if your company has 1,000 employees and only one person actually wants to use the AI features, you still have to pay for all 1,000 seats. What they do not seem to understand is that these days, data does not necessarily need to live inside a tool like Notion. More and more, it feels like data will end up living in LLM-based systems instead. As someone who has used Notion for a long time and has been a paid subscriber for more than five years, I really think Notion needs to rethink its AI strategy carefully.

@WisemanCap I am on claude/cursor bro. Moved away from Google AI, despite having been a huge fan. I only use Google One 2TB for storing photos and stuff. I don't use any of their AI features.. since it is limited to google ecosystem while claude/cursor gives a whole range option.

So Google has upped the storage to 5TB for Pro plans!

@jiggycapital We call it Skill 🤷‍♂️

AGI is truly a markdown file

@convequity Prob embeeding still has decent play not from the pov you are showing but on the backend for RAG.

Vanilla markdown file-based memory system is killing vector store. OpenClaw, Claude Code, Manus, and other successful agent architectures have greatly amplified base model's performance via tiered short-term, near-term, and long term memory structure. This is so funny, and, again, the most ironical outcome is the most likely outcome. Previously, the consensus was that RAG-based on embedding/similarity-based search results feeding to the input prompt. Now, it seems like what works the best for the model itself is to keep things simple, let models figure out things by themselves, and let models write memories. Incumbent software companies spent too much time engineering a sophisticated RAG agentic system that is now quickly rendered obsolete by new approaches by newcomers. Again, this further supports the narrative that in the AI era, incumbents are more vulnerable vs. startups. Interetsingly, $META spent $2bn for Manus. Compared to OpenClaw, Manus has a more sophisticated isolated computer-use environment and harness engineering. But the risk remains - can Meta grow the Manus asset or will Manus be dragged into corporate politics and being written off as well? Also, what this architecture entails is that maybe we don't need highly curated structural data like data warehouse or even data lake. But maybe if we do believe that LLM is the virtual human brain, then just like our memory system. As human beings, we don't really play with database directly and instead we rely on scratches, notes, and loose plain text to augment our memories. That also makes me wonder - does system of record really matter? Maybe yes over the short term, but over the longer term, a close to plain text LLM-native memory structure for SOR is also needed? If you need to extract and transform the data via an efficient small model out of the SOR, then SOR's lock-in doesn't really matter that much.

@YingjunWu you are living in a tech world and 80% of the country prob don't even know what github is but def know what Tiktok and IG is.

1K+ stars in 5 hours... I've told you that GitHub has become the new TikTok...

@lefttailguy Not in my house-hold. I am a huge supporter of proper 6-8 hours sleep. Kids these days abuse their time. It has nothing to do with parents demanding more time.. They want kids to manage their time better.

@sv_techie I think parents would disagree here.

Is anyone working on a way to make us all short sleepers (DEC2 / ADRB1 gene)? The benefits to productivity and economic growth would be profound. Imagine if everyone on earth had 4 more productive hours per day.

@lefttailguy that will take a long time unless we naturally evolve.

@sv_techie I'm talking about a way to shorten sleep without the negative effects.

@lefttailguy People already have plenty of hours even without touching 6-8 hours sleep - you don’t really need to muck around with sleep time fwiw.

@lefttailguy Altering biological needs has its own set of challenges - sleep is fundamental to mental and physical health.

@thingsnoticed Yes - def many unknowns. But next few years is going to be rough is what I feel. We haven’t had a real R in a while.

@sv_techie thesis assumes labor redeploys history supports it but AI is horizontal so where does new demand come from if all industries need fewer people?

It is becoming very clear, what you can do with claude, cursor, etc., is amazing from productivity pov especially w/PDLC of SW. What needed 5-10 people is now 1-2 people in many places. AI spend will be funded by having less humans in the near term (i.e layoffs).

@thingsnoticed it should reduce labor but then if we are able to deploy surplus to create more then productivity boost even further but that requires retraining folks, etc., will take some time to redeploy.

@sv_techie does AI productivity reduce labor or just change where it’s needed?

@jake_researcher Agree - def quality is a huge concern. Also it is not about just coding - lot of design, systems thinking, security / guard rails, etc., But def there is a lot of benefits in the hands of very smart seasoned people.

@sv_techie The 1-2 person team shift is real but I wonder about the quality ceiling. When something needs 10 people, it's often not just capacity but diverse perspectives catching issues. What happens when that synthesis disappears?