Mudit Juneja (@being_mudit) profile picture

Building a lot of things but succeeding only at sarcasm.

New Delhi, India
Joined June 2015
Photo by Mudit Juneja (@being_mudit)Photo by Mudit Juneja (@being_mudit)Photo by Mudit Juneja (@being_mudit)Photo by Mudit Juneja (@being_mudit)Photo by Mudit Juneja (@being_mudit)Photo by Mudit Juneja (@being_mudit)Photo by Mudit Juneja (@being_mudit)Photo by Mudit Juneja (@being_mudit)Photo by Mudit Juneja (@being_mudit)
Banks lose millions to database race conditions. TigerBeetle solves this with built-in double-entry accounting primitives that make financial bugs mathematically impossible. Where Postgres needs complex locks, this just works.
24
46
6
706
68,245
Mudit Juneja retweeted
I am helping @getplumhq build b2c health insurance experience , it is a 0 to 1 problem statement for which we are hiring a backend engineer; what I promise: high autonomy to experiment + end-to-end ownership RT for good karma 🙏🏻 careers.kula.ai/plumhq/59448…
1
19
2
71
4,825
Time for Ultron level personalised intelligence.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
21
Mudit Juneja retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2,358
7,273
5,720
55,927
9,796,553
Mudit Juneja retweeted
A few weeks ago…
I've done a complete U-turn on my opinion on open source AI. We should be careful (and probably disallow release) of models on par with current open source Astra/Opus level models. AI is now reverse engineering games from binaries and remaking them. Reverse engineering games is a very hard problem. If this is possible then reverse engineering banking software, ID systems (Aadhaar), flights, etc is possible too. A game ships its binary to every player. Bank and Aadhaar backends are protected by servers, but it's unlikely the security teams at these are smarter than advanced AI that can do this to games. We are protected rn because a Claude will refuse to tick "I'm not a robot" on websites. When OSS models don't respect that, we have a looming cybersecurity problem. Apologies I didn't see this earlier.
27
4
275
19,356
I guess MailCheap is not that cheap. Ahh the irony. What matters most for indie devs/vibe coders is pay wayy less than $19 for an email plane and still have flexibility for expanding as their user base grows. Why pay for enormous email capacity when you have just a couple of users. Not all of us are blessed like Marc himself. @sendwithxmit (Transmit) bridges that gap and have a $2 plan for the transactional route. Or you can bring your own key with AWS SES with the free plan with all agentic capabilities.
I vibe-coded beehiiv, and it's 83% cheaper. ✨✨ Introducing MailCheap.ai ✨✨ Everything you need to run a newsletter... ✍️ Editor 📅 Scheduling 📈 Stats 🌐 A website for every issue 🛡️ Deliverability guardrails 🔗 Send from your own domain (and keep your reputation) And it's not a dinosaur tool: + It's open source; you can run it for free + It's agentic; everything in the UI is available to your AI Import from beehiiv, Substack, or Mailchimp in 2 minutes. You pay for the emails you send, not for dormant subscribers. From $19/month. Oh, and no "Powered By" badge at the bottom of your emails.
1
1
111
“I just want magic link emails + weekly newsletter AND NOT PAY $49/MO” This is the actual majority use case. Transmit (@sendwithxmit ) was built for exactly that. Transactional + campaigns, domain warmup, suppressions, and lists, without the full beehiiv surface area. Managed starts at $2.
ppl dont realise that i just want to send magic link emails + weekly newsletter AND NOT PAY $49/MO i am not competing with you i just dont need 99% of features u have
25
This is exactly why volume-based pricing only works if the rates (and infrastructure) are actually competitive at scale. For high-frequency newsletters, pure “pay per email” can get expensive fast - unless you have proper BYOK / SES rates + solid deliverability. Worth looking at @sendwithxmit (xmit.sh) if you’re sending serious volume and care about inbox placement + isolation.
lol so here’s what the headline doesn’t show he’s charging per email sent not per contact. meanwhile @beehiiv lets you send unlimited times per month for the same price I have a 50k subscriber newsletter that I send daily. That’s 1.5M emails a month. That’s $349 on this which is literally more than beehiiv 😭
3
64
Mudit Juneja retweeted
I'm surprised by how good AI music generation has become. It's almost become indistinguishable from amateur productions. Presenting R*tard in Motion (watch with sound: ON) I've been exploring how to produce music and video using AI and decided to make my own Rap music video. I always really enjoyed the discourse between "Idiots in motion vs Genius at Rest" and turned it into a song. Considering how I made this in a few prompts and half a day, it's safe to say the AI race is just getting started.
32
19
7
227
18,368
Mudit Juneja retweeted
As someone who has lost family to cancer, this is an easy answer for me. If something has passed clinical trials and could potentially prolong the life of someone dying I'd take the risk. I can always rely on someone to understand how the medicine works. And would even sign waivers to let them examine and test the person it was administered to so they can study it. I think we underestimate how desperate people become when it comes to their loved ones.
This is a slide from Terence Tao's 'Math 2.0' public lecture yesterday evening at Caltech. I am highly confident that, if the cure actually worked, almost any Stage 4 cancer patient would tell you that human comprehension of its mechanism is completely irrelevant to them.
3
2
29
914
Mudit Juneja retweeted
we're approaching 20M monthly active users there's maybe 50-100M developers in the world so either 1 in 5 of them use OpenCode or.....
twitter is a bubble normal people are not vibe coding their own software calm down
203
34
11
3,120
256,085
The irony in people’s life. His entire empire seems AI generated like 99% and what if his users start saying this. Wth is wrong with the world.
37
Yep Opus 5.5 did really great! Product demos are no longer tough no longer. It helped me create a promo for Koin in just under 12 dollars.
Holy shit. Just 1 year ago, I would have to paid ~$1,000+ for a video like this. Now I made this with Opus 5.5 in Cursor within 1 hour, one shot.
1
1
2
587
Mudit Juneja retweeted
Holy shit. Just 1 year ago, I would have to paid ~$1,000+ for a video like this. Now I made this with Opus 5.5 in Cursor within 1 hour, one shot.
4
1
3
26
4,803
This is needed every where in India. @RampurDm i would like to help with a prototype if Rampur is committed.
Gajendra Jangid, co-founder of Cars24, just committed to bringing or building an instant pothole fixer for India. fyi @gajen_jangid the Python 5000 in this vid (shot in Ottawa) is made by Regina-based @SuperiorRoads. C$400k (₹2.71 crore)/unit. Likely cheaper to make in India!
1
4
151
This is funny on so many levels. @AnthropicAI is savage 😆
37
Mudit Juneja retweeted
Let's go local AI! At a time when everyone is complaining about shortages of energy and compute for AI & costs of tokens, we need on-device AI more than ever! In addition, it's virtually free & fully privacy preserving!
Today we're launching Desert Ant Labs: a European frontier AI lab building on-device intelligence. 18 models across audio, vision, and text. SDKs for Swift, Kotlin, and JavaScript. No tokens. No logins. Nothing leaves the device. desertant.com
91
180
23
2,633
279,593
Holy Moly! This is crazy to just see. Apple can’t be doing this that easily @dhh what did you have to do?
The day you give up on your dreams, you might as well just give up! (Look under OS 😂)
56
Mudit Juneja retweeted
For those who want to understand how this works: 1) They open a webview with the site they want you to login to 2) You login to that site, creating some cookies 3) They transfer those cookies to their agent’s computer 4) Now your agent can access that site Surprised no one did this before them TBH. It’s great because it still allows your password manager to work since the authentication happens on your device. I wonder what the story around at rest and in transit encryption is.
We built the most secure, simple way to sign an agent into websites from your phone. The agent never receives your password. Not in the chat, not in the logs. This works across all sites, and any sign in patterns.
89
39
4
1,214
223,743
All it takes is one clear six-minute explanatory video on how the glacial collapse happened to tell you that pretty much a lot of what's been said on Indian media and social media these past few days has been complete misinformation.
22
206
10
747
55,698