Chintan Goyal

Promoted to a Staff Engineer ! Came from Avg college, mechanical degree and now earned the respect and recognition from leads and peers from MAANG and the top tier startups across global teams. The journey is nothing but fun and I'm excited for what's coming next !. LFG !!!

Why Google IO looks like a Teenager's Live stream with all the gameplay

@antigravity @GoogleDeepMind Please use Google AI Studio and build better app for antigravity. Let me know if you need helping hands here :)

The only beef which I have with Codex is 5h limit. It breaks focus and concentration if I'm working on something important. Weekly limits or daily limits are fine, but five hour limit is very disturbing. @OpenAI @sama

Whisperflow actually created the whole market and made people realise that speech‑to‑text can be accurate and cool, People have realized that it's very easy to build, too. Now everyone is building voice‑to‑text for phones and desktop. 🙃

i built something cool! steno is a lightning fast on-device transcription app for the iphone. you can swap it out for your keyboard in pretty much every app and it just works. had a lot of fun designing the interface. link below if you're interested -- shipping soon :)

@archiexzzz @sugandhanisa Check Voquill and VoiceInk

no disrespect to him or WisprFlow, but i was able to match the same accuracy and latency as WisprFlow in 2hrs of working on something similar. have many closed-beta users using the product and most have already replaced wisprflow with my product. but i feel voice-to-action is a bigger market

Tanay Kothari is a super strong founder, but whoever is ghostwriting for him needs to be fired before his slop that always ends with "written with wispr flow" gets recognised by the average LinkedIn user. Also, what the hell do you mean that pre-AI you would have needed a full ops hire, a sourcing agent, a designer for ordering 100 BEDSHEETS? Is this how we are peddling the AI hype -- by claiming that whatever AI agents are doing would've needed 5 full-time hires earlier, even when the topic is ordering bedsheets? Come on.

It is quite interesting that from last 3 months, there is an underlying pattern in the AI ecosystem, if you notice, you will be able to read between the lines and cut through the actual reality in the fog of AI psychosis and the marketing hype > Till Dec 2025, New Model releases, bring genuinely improved intelligence > Cost remains nearly same, $20 gives immense usage, $200 is luxury. > Opus 4.5 launches, peak time 2025 Dec, everyone goes mad by seeing capability of the model. > Slowly, the limits starts drying up ( Feb ), > March, single prompt of Opus is now taking $20 > Anthropic nerfs it to ~65%, user finds out immediately, they took L . > GPT 4.5 and Codex 5.3 starts getting attention, $20 gives you good enough usage. > ChatGpt comes back in the race , people shifts, load grows, they cut the usage limits. > Opus is unusable, Codex limit is reached, API bills starts increasing rapidly across services and each company starts increasing price + cutting limits. > Mythos teaser drops, next day, people and community debunks the hype and the hype dies immediately. > Anthropic finally accepts the issue, try to bring back 4.6, but by the time, the damage is done, devs leave > They release 4.7, which comes out to be costly and worse than 4.6. > Billing is going up, limits are going down. > Reliability and trust on the private model providers is gone, seeing how easily they can nerf a model. > After 3 years, they figured out that the only market AI can be used to earn money is developer productivity + bit automation with chatbots. > Both requires reliability + cost efficiency, both are gone in Start of Q1, 2026

Hi @bcherny , @ClaudeDevs , @claudeai This is Opus 4.7 :) . This is supposed to take software engineering jobs?

@MarioBalukcic @9to5mac once you experience Apple refinements, there is no going back

@9to5mac I’m don’t plan on switching back to Android. iPhone just works.

iPhone brand loyalty just hit a record high, according to a new large-scale survey of US smartphone users. Loyalty climbed from 90.5% in 2019 to 96.4% in 2026. Android users are also reportedly switching over at nearly 4x the rate.

@synthwavedd Any decent human engineer would've tested it better, seems like they completely vibecoded and pushed it, maybe for a case study for Mythos, but turned out pretty poorly :)

if you wanted proof mythos is far from AGI, this is it

Enough time has passed, but @browsercompany is still not accepting that the only way to build or make @diabrowser better is to build it on top of Arc. If they want, they can have all the MVPs from it, but it seems that the fundamental design is different. Dia feels like another Chrome or Comet, but their original product Arc is something completely unique and fresh ( Which, even after being abandoned, is by far the best browser in terms of refinement and ergonomics ) Hope @browsercompany comes to the senses soon, and if they don’t want, then just open source the browser, community would be more than happy to have it

Romanised Hindi ( Hinglish ) is the primary way in which wider Hindi speaking population interacts online. In my experience Every top STT ( dictation) app is highly inconsistent at it, just because it is a mix of English + Hindi in roman script. researched and specifically used an opensourced STT Oriserve/Whisper-Hindi2Hinglish-Apex, Ran it locally, used it with Voquill and bang, And this is already doing the best in class on-device transcription. Example below:

This weekend I was trying to understand the whole transcription thing. Majorly it is a STT transcription model and post processing problem specifically with Romanized Hindi. I as per my observation, Whisperflow is using the open source Whisper V3 large model. I found very close matches and then they are doing some post processing. But I think this problem could be better solved by Sarvam stt

@gchintn @josiahsrc @WisprFlow thing might be a memory problem

how hard is it to get one name right?? i've added diwank to @WisprFlow 's dictionary so many times and it still keeps finding new ways to miss it.

@IshitaJindal17 @josiahsrc @WisprFlow Not exactly sure, but we would love if you can create a PR and solve it for all. Also, we can make Hinglish better there.

@josiahsrc @WisprFlow i know. would voquill's dictionary work better?

@reactive_dude We have been struggling for the same since last two weeks and now it is clear that both of them are not reliable since they can make the model dumber any day and choke us with rate limits. A reliable open source deployment seems to be the way from here

I've been complaining about Opus being weird today...and people told me to use Codex. And now all my timeline is people complaining their rate limit on Codex is messed up? Are you kidding me? What is going on? lol

Found out about Voquill , the person and community who is building a completely open sourced and free Wisperflow alternative. Absolutely great product and vision and I feel Obsidian kind of revolution :)

@NotebookLM Generate full length audiobook from a document

We intentionally didn't post anything over the past couple of days as to not abuse your trust nor insult your intelligence 😉 However... If we *had* wanted to fool you, what were you hoping to see? (Think beyond the usual requests! What's your dream feature request?)