Harj Taggar (@harjtaggar) header
Harj Taggar (@harjtaggar) profile picture
Joined October 2006
Photo by Harj Taggar (@harjtaggar)Photo by Harj Taggar (@harjtaggar)Photo by Harj Taggar (@harjtaggar)Photo by Harj Taggar (@harjtaggar)Photo by Harj Taggar (@harjtaggar)Photo by Harj Taggar (@harjtaggar)Photo by Harj Taggar (@harjtaggar)Photo by Harj Taggar (@harjtaggar)Photo by Harj Taggar (@harjtaggar)
Harj Taggar retweeted
Strong American open source models are needed to make the US supply chain more resilient and to strengthen the economy.
Super Intelligence Force Vice Chair Emil Michael (@USWREMichael) on the need to proliferate American open source SI models: "NVIDIA has been a huge leader in American open source... we really need American open source to win... I hope whoever leads NVIDIA continues to carry that mantle of promoting American open source, because that's also a defense against too much concentration in American closed source. That's really something I hope to see from the company in the coming months."
15
12
3
234
15,179
I've been working with Henrijs this batch and he's a force of nature. He's been zoom'ing into office hours from Ukraine.
Replying to @WIRED
In September, Dandzbergs, who was born in Latvia, was admitted to Y Combinator’s fall 2026 cohort. The startup accelerator counts Airbnb, Stripe, DoorDash, and Reddit among its alumni; Dandzbergs is likely the youngest solo founder in its vaunted history. wired.com/story/henrijs-dand…
1
36
7,421
Harj Taggar retweeted
I caught up with a YC founder in this batch who I met right before the batch. The difference between then and now is genuinely startling. The confidence, the way he's talking about the vision, GTM, hiring, fundraising. Felt like a different person altogether. Even though I've been in the yc ecosystem for 8+ years now, it still surprises me how much @ycombinator can change the trajectory of a founder and company in a few short months
9
6
137
11,682
I was very impressed with @slush. It might seem a crazy idea to invite people to Finland in November, but it worked. I had a great conversation with @antonosika about the present and future of AI.
16
17
1
265
12,826
Harj Taggar retweeted
we've made sure you can use the opencode console with any harness this means you can use any model available in opencode from any popular harness while keeping console features like budgets or usage tracking because you deserve the best inference even if you make bad decisions
18
23
1
452
14,448
Harj Taggar retweeted
Idk, if you don't use the models for coding or data-oriented tasks, you might think this. And I don't think super-skilled coding = super-intelligence. But if you do, it's obvious the models are pretty fucking smart and rapidly improving.
I think AI models have probably reached a permanent plateau. Have seen little improvement in last 9 months. RSI is a more complicated problem than people think: if models have natural limits there is no RSI that can overcome them
76
41
14
1,867
177,206
Your workout is my warm up
This quoted post is unavailable.
2
17
1,478
Harj Taggar retweeted
One of the few delights of having leftists in power is watching everyone discover their thought process. This is how leftists think and speak, and whenever they leave their intellectual bubbles they fall apart with the slightest bit of pushback.
207
525
33
6,730
567,771
Harj Taggar retweeted
He's nothing like this. The tech industry is nothing like this. A freakish 'everyone who's rich is dumb and weird and evil' fantasy for the groundlings.
New trailer for Luca Guadagnino's ‘ARTIFICIAL’, starring Andrew Garfield as Sam Altman. The film follows the story involving OpenAI focused on the firing & rehiring of CEO Sam Altman. In theaters on January 8.
208
136
42
3,872
267,923
Harj Taggar retweeted
🚀 Starting today, @Rippling AI can build sophisticated business tools using your company's data. Any user can prompt their way to useful, beautiful business software. Here's an example showing a recruiting pipeline, but let your imagination run wild... 🧵
2
23
4
131
33,010
True for building all good products
You will notice that Gemini Carbon thinks differently than Claude and GPT. It often comes up with entirely different but more clever solutions than competitor models. This is precisely because we never distilled other models. Distilling other models creates a poor imitation rather than unique intelligence.
8
2
102
11,294
I want to thanks @SemiAnalysis_ for the motivation. In the Gemini RL scaling team, we printed & put that article on our desks to be reminded daily of what used to be true at the time. But the times they are changing :)
Gemini is Cooked but GCP is Cooking GCP YoY rev growth >100%, DeepMind's long term failure is Google Cloud's short term gain newsletter.semianalysis.com/…
149
277
93
5,010
547,123
Harj Taggar retweeted
this guy is an insane marketer
🚨 MUSE FEATURE ALERT 🚨 you can now try on clothes in muse before you buy them 👕 👗 👖 muse will offer to show you how you look in something while you’re shopping for it. trying things on in person? as if....
51
56
8
2,889
528,124
Harj Taggar retweeted
I did YC in W20 and I'm back for F26. The software YC has built since then is wild. In 2020 I was dragging SAFEs into DocuSign and placing signature boxes by hand. Now it's one click: the SAFE gets signed and wire details go out automatically. Investor CRM, investor calendar, Granola integration, live pitch feedback, all built in-house. But the best part is warm intro paths. YC has turned its community into a real network, and some days it feels like the best (only?) way to build a startup.
16
9
3
117
18,632
Harj Taggar retweeted
reflecting on this, it was in august where deepseek 4.1's success and price hike destroyed our economics we talked to every inference company that existed. startups and massive companies as well none of them could deliver what we needed. we owned no hardware. had zero leases. had never managed a GPU directly and 3 months later we got GPUs melting and serving deepseek and glm at the exact same quality as the labs that make them if you're a hater (or worse a ceo that's likes to subtweet us) find a new target unless you particularly enjoy pain
44% of our tokens today were served by our own GPUs that's 16T tokens and we have 10x the hardware coming online in the next few months let's go
120
91
19
3,878
146,049
Harj Taggar retweeted
44% of our tokens today were served by our own GPUs that's 16T tokens and we have 10x the hardware coming online in the next few months let's go
143
136
35
4,104
272,250
Harj Taggar retweeted
This is a product I've wanted for a while: a shared filesystem for people and their agents. It's a simple idea that solves so many problems with agent work.
Today we’re launching FlashDrive with @ycombinator 🚀 FlashDrive is a shared cloud drive where your team and all your AI agents work in the same files across computers and cloud sandboxes. > Search the whole drive with grep, regex or code. Our custom indexing layer finds matches without downloading everything. > Files stay in the cloud, and your agent only streams the part it needs. > Start in Claude Code on your Mac. Pick up in Codex. > Edit the same file at the same time. FlashDrive merges changes or keeps both versions. > Connect over MCP or mount the drive on macOS or Linux. Give a sandbox access to just one folder for as little as five minutes. Dalton and I are building the filesystem the next generation of companies will run on. And we're calling it @flashdrive_ai ↓ Try it free! Link below
37
26
1
507
72,547
GTA 6 is so realistic that a VC tells Jason "love what you're building, let's catch up next week" and then disappears for the rest of the game.
17
63
7
2,103
60,278