X Stats
Login Sign Up
@levelsio

@levelsio

@levelsio

810 Tweets
2,563 Following
790,172 Followers

🇪🇺https://t.co/NdorAWqJC3 @euacc 📸https://t.co/lAyoqmSBRX $110K/m 🏡https://t.co/1oqUgfD6CZ $35K/m 🛰https://t.co/ZHSvI2wjyW $43K/m 🌍https://t.co/UXK5AFqCaQ $17K/m 👙https://t.co/RyXpqGuFM3 $14K/m 💾https://t.co/M1hEUBAynC $5K/m

Tweets Topics Chat
Filtered by topic: AI Chatbots & LLMs ✕ Clear filter

Filter by views

All 810 < 500 133 500 - 5k 0 5k+ 677
Date ↓ Tweet ❤️ 🔄 💬 👁 📝
Dec 27 Google's Nano Banana Pro is expensive but also not They "only" need to generate about 300 million to 1 billion images to turn a profit! https://t.co/7CtOPr08VZ 49 0 9 27.2k 161 .
Dec 27 Very good question Grok estimates 🍌 Nano Banana Pro cost Google about $50-$200 million just to train Every year the models get better (and bigger), and generally require more compute (like others have said), so the next better model will cost a multiple of that, so like $500 million or even $1 billion So unless I'd raise $500 million to train that next model I don't think it's possible, because I simply do not have $500 million lying around Another thing is that it's starting to become clear being in the model business is generally a very bad idea for most people because the moment a better model arrives, everyone just changes their code: from: > 'model'=>'current-model' to: > 'model'=>'the-new-model' As in, the switching cost is almost $0, so models do not have much lock in Which is WHY I think you'll see the best models come out of BigTech, because they can afford to spend massive money on creating the best models, and then make that money back by using those models in their entire portfolio of apps (see Google) I don't think long term as a AI model startup you can keep up with that, because they 1) can't raise enough money for the trainings to compete with BigTech's infinite pockets, 2) can't survive the moment a better model comes around, 3) they don't have the front end and users to get a moat What that means for my AI apps though is that it might be better I don't have my own models because I don't face the financial risk of training a model that's outdated in 6 months, and I just keep updating to whatever new model comes out (yes like a wrapper) Being a niche AI app like my app Photo AI is, I can actually offer features that make it better than a general AI experience you'd have on ChatGPT or Gemini so it can remain a long-term business that way by specializing on a specific area of people photography 319 14 58 74.6k 1.8k .
Dec 27 I lose money on every image, 🍌 Nano Banana Pro is too expensive But the quality is just too good, so I hope to get the profit back on other features 638 9 86 141k 149 .
Dec 21 You can try it here https://t.co/RRYOCWrpFY 55 2 3 31.8k 43 .
Dec 21 Okay I improved the chat a bit on both AIM and IRC, it has more tools now like Search and Thinking so you can actually have deep convos with it now that can query the web, very nice I tried asking it what's going on on Hacker News today and it told me Chat Control 3 and I asked about what it was etc. and what it thought about it, all inside AOL Instant Messenger 😊 Feels really fun to have access to modern LLMs on retro Windows 197 5 25 148k 432 .
Dec 21 Interesting story An AIM chat bot around 2000 that was actually like a ChatGPT back then, written in Perl!! https://t.co/jSsZ0pu1tM 310 8 35 103.9k 132 .
Dec 21 💾 Okay having nobody online to chat on AIM is boring So I asked AI to write an AOL Instant Messenger bot called @pieterbot, and I made an account for it, and it one-shotted it in Python and IT WORKS!!! So now you can chat on AIM on https://t.co/M1hEUBAynC with an LLM that is fully self-aware it is an AIM bot 🤠👍 327 10 52 273.9k 314 .