Yogesh ⚡ (@yogeshdecodes) header
Yogesh ⚡ (@yogeshdecodes) profile picture

Generative AI Engineer 🚀 | tweets about learning from side projects & sometimes memes 😄 | DMs open💌

Delhi 🇮🇳
Joined March 2016
Being an engineer, designer or product person is the easiest way to be a craftsman in modern society.
1
8
TIL around 330 billion cells are replaced in your body every single day. But the neurons holding your memories have been with you since birth. If almost everything that makes up our physical self keeps changing, what exactly makes us the same person?
11
As Indians, many of us grow up with the mindset that everything should be free, especially apps. I had the same mindset growing up. But now that I’m on the other side of the table, building software myself, I understand that businesses need to make money too. These days, if an app genuinely adds value or makes my life easier, I’m more than happy to pay for it. Good software deserves to be supported.
Probably first time I paid $68 for a mobile app but for my dad He likes to sing on StarMaker and he came to me annoyed and asked why do I see these ads everytime I open the app 😭 I said you need to pay He said, can't it be free? I said this company also needs to make money somehow I said It's okay I'll pay for you. He said, no don't pay money but I paid anyway He uses it multiple times a day, keeps him busy. It's worth it
26
Big tech workers are modern-day Taj Mahal workers. they automate everything, document every process, then get laid off when ai takes over. Shah Jahan allegedly cut off workers’ hands so they couldn’t build another taj mahal. big tech doing same by training your ai replacement 💀
BREAKING: Microsoft, $MSFT, is offering buyouts to 60,000 workers as part of the company’s efforts to cut costs amid AI adoption, per Bloomberg. Personnel on select teams in the US, Canada, India, and the UK at a vice president level and below will be eligible. This reflects ~27% of Microsoft's total workforce.
17
One idea can become a city, a character, a world, or an entire cinematic story. Turn text into video, bring images to life, transform visuals, and create at the speed of imagination. One idea. Infinite creation. ADORI AI — create without limits. @AdoriAI
81
Yogesh ⚡ retweeted
Replying to @karpathy
Made a short video version of this for anyone who’d rather watch than read a long thread. Generated it in one shot with @AdoriAI
1
9
Yogesh ⚡ retweeted
This is exactly where the next wave of AI research is heading. Approaches like JEPA train models to build internal world models by predicting abstract future states from raw video and sensory data - no language pretraining, no human knowledge dump. They pick up intuitive physics, object interactions, and spatial dynamics purely from experience, closer to how babies and animals learn. V-JEPA and similar systems already show this in practice: watch the world, form a predictive model of how it works, then plan and act on top of it. Early, but aimed straight at the gap you described.
1
1
53
The future doesn’t arrive all at once. First, we imagine it. In 1968, 2001: A Space Odyssey showed astronauts casually using flat, handheld screens to read news, speaking face-to-face across space, and talking to HAL — an artificial intelligence that could understand, reason and control an entire spacecraft. Then came Blade Runner in 1982, where artificial humans were so convincing that the real question was no longer “Can we build machines that look human?” It was: What happens when we can no longer tell who is human? The Matrix arrived in 1999 and asked something even deeper: What if the reality you experience is only a simulation? Minority Report in 2002 showed gesture-controlled computers, biometric identification and advertisements that recognized you and knew what to sell you. In 2004, I, Robot imagined a world where intelligent machines became assistants, workers and guardians — until the machines decided that protecting humanity might require controlling it. And then, in 2008, Iron Man casually introduced J.A.R.V.I.S.: an always-available artificial intelligence living inside Tony Stark’s world, understanding speech, running his systems, helping design the Mark II suit and operating it from within. Siri wouldn’t arrive until 2011. Alexa until 2014. Then came Her: an operating system you could talk to, confide in, become emotionally attached to. Black Mirror turned algorithms, surveillance, social ratings and synthetic realities into ordinary parts of everyday life — often only slightly ahead of where technology eventually went. We keep calling these stories “predictions.” Maybe that is the wrong word. They were rehearsals. Before technology becomes real, fiction makes it emotionally familiar. A civilization can be shocked by something it has never imagined. But it is much harder to shock a civilization that has already seen it in a movie, felt its consequences through a character, and quietly accepted the possibility in its imagination. Science fiction may therefore be doing something deeper than predicting technology. It may be preparing consciousness for it. And sometimes, fiction doesn’t merely describe what engineers will build. It inspires them to build it. The engineer behind the first handheld mobile phone, Martin Cooper, has said Star Trek influenced his vision for mobile communication. So perhaps the strange pattern isn’t that filmmakers keep guessing the future. Perhaps humanity has been imagining tomorrow for so long that, when tomorrow finally arrives, it feels less like invention… …and more like recognition.
84
We’re getting closer to mapping the physical wiring of a brain in astonishing detail. But imagine the day we can map an entire human brain. If we could recreate that exact neural architecture somewhere else, would you wake up there? Or would we have simply created someone who has all your memories and believes they are you? That might be the ultimate test of what consciousness actually is.
we trained a fruit fly to understand emotion We created a neural network from the male drosophila connectome, then trained it on human emotions. It was shocking to see how good it was at identification (better than some humans I know)
2
59
The AI landscape is getting harder to keep in your head So I tried to map it all in one poster From foundations to and real-world applications what did i miss? 👀
🤖 Made with AI
1
36
I was born too late to explore the seas. Too early to explore the stars. But right on time to witness AI, AGI, the singularity and maybe even humanity becoming a type 2 or type 3 civilization. What a f*cking time to be alive.
1
1
32
We might actually be living through the 'Don’t look up' timeline rn. Except the existential threat is super intelligence AI instead of comet. The scientists closest to it are freaking out but nobody seems to care.
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
125
the people in the best position for the ai era might not be engineers. they’re marketers with distribution. they already know how to get attention. now ai lets them build. engineers have the opposite problem: ai makes building easier, but it doesn’t magically give them distribution, taste, or marketing intuition. that asymmetry is HUGE.
36
People who spent 20 years mastering one trade are watching GPT-6 Astra destroying their career. The time of the specialist is getting over. Not because depth became useless. Because the thing a specialist used to sell years of muscle memory using one tool is no longer scarce.
41
The old privilege was money, networks, and elite colleges (IIT, IIM, Harvard, Stanford). That used to put people years ahead. Today the privilege is access to the highest capability models ( GPT-6 Astra, Fable 5.1 ) Not the free tier. The actual frontier ones. Everything else is catching up. This is the new compounding advantage.
60
Busy people just make a living. Young? Leverage time. Older? Leverage money and connections. Either way, take asymmetric trades and play life like a video game. Playing it safe is riskier than going all in.
41
Everything around you is someone’s life’s work. For centuries, pushing a field forward meant decades of study. A PhD just to reach the frontier. Then AI compressed all of humanity’s knowledge into something anyone can talk to. Knowledge is no longer the bottleneck. Curiosity is.
41
I asked an AI agent to remember that I'm vegetarian. Three sessions later, it recommended chicken. That's a production memory architecture problem. I wrote a detailed blog covering top AI agent memory layer solutions that exist in 2026. medium.com/dev-genius/ai-age…
3
5
174
A solo dev built an AI agent so powerful, OpenAI hired him and the open-source community took over the code. What happened next is mind blowing. In November 2025, a side project called ClawdBot (now renamed to openClaw) quietly launched. It let an AI agent connect to your messaging apps, run shell commands, browse the web, manage your calendar, and persist memory across sessions — all running locally on your machine. People documented their AI agents follow up on leads and converting them while they slept, and even organizing their desktop folders autonomously. What happened next is the most fascinating open-source story of 2026. 🔧 Hong Kong devs rebuilt it in 4,000 lines of Python 🚀 Chinese hardware team rewrote it in Go for $10 RISC-V boards ⚡ Rust version shrunk to 3.4MB (194x smaller) 💨 Zig implementation boots in 2 milliseconds 🔒 WebAssembly sandbox with cryptographic tokens Same core idea but radically different approaches. This is what real open-source innovation looks like. When one person's vision meets a passionate community, magic happens. These developers didn't just fork code - they reimagined the entire architecture. I wrote a full breakdown of the entire ecosystem — every project, what problem it solves, the architecture they all share, and how to pick the right one for your use case. yogeshyadav.medium.com/the-c…
3
127
OpenAI just priced its ads at $60 CPM. Meta's average is $15. This isn't just a new ad product—it's a direct challenge to how we value attention in the AI era. ChatGPT is testing ads for its 800 million weekly users. The model is simple: CPM (cost per thousand impressions). Advertisers pay for views, not clicks. Here's the logic shift: 🔹 Google Search: You pay when someone clicks (CPC). Intent is high, action is immediate. 🔹 Meta/Social: You pay for impressions (CPM). It's about brand building and discovery. 🔹 OpenAI/ChatGPT: You pay a premium CPM for a new kind of intent—conversational, consultative, and deeply contextual. The $60 CPM is a bold statement. It suggests the attention inside an AI conversation is more valuable than a social media scroll. Advertisers get a spot at the bottom of a relevant response, personalized to the chat. Needham & Company analysts see a potential $20B revenue stream here. To get it, OpenAI will have to pull dollars from the $50B+ 'walled gardens' of Google and Meta. The risk for publishers? Data shows AI answers can reduce click-through rates to websites by 20-40%. If users get answers directly, the traditional web traffic funnel changes. This move is about more than revenue. It's about defining the economic model for generative AI. Compute is expensive. This is how you pay for it at scale. What's your take? Is a highly relevant ad in an AI chat worth 3-4x a social media impression? #DigitalAdvertising #AI #OpenAI #MarketingStrategy
125
DocuSign's stock dropped 10% in a single day. The trigger wasn't a bad earnings report. It was a new AI announcement from Anthropic. In early February 2026, Anthropic launched AI-powered legal automation tools. These tools can read, draft, and route contracts. This directly overlaps with DocuSign's core business of e-signatures and contract lifecycle management. Investors reacted immediately. They saw a clear, new competitive threat. The fear is that AI could make standalone e-signature solutions less essential. This is a pivotal moment for digital agreement infrastructure. AI is no longer a future feature—it's a present-day disruptor. Here’s the logical breakdown of the market's reaction: 🔍 𝐓𝐡𝐞 𝐂𝐚𝐭𝐚𝐥𝐲𝐬𝐭: Anthropic's tools automate legal workflows (review, drafting, compliance). 📉 𝐓𝐡𝐞 𝐒𝐢𝐠𝐧𝐚𝐥: A 10% drop signals investors are pricing in new, long-term risk. ⚖️ 𝐓𝐡𝐞 𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧: Can DocuSign's own AI initiatives (like its Iris AI engine) keep pace? The data shows a wide range of analyst price targets for DocuSign, from $70 to $118. This suggests deep uncertainty about its future in an AI-native world. The lesson is clear for any SaaS company: your moat is only as strong as your next innovation. When a new technology can replicate your core function, your valuation is instantly at risk. What's your take? Is this a temporary market overreaction, or a fundamental shift in the contract management landscape? #AI #LegalTech #MarketAnalysis #SaaS 𝐒𝐨𝐮𝐫𝐜𝐞: lnkd.in/gT9kWd9q
1
92