Tencent launched Hy4 flash today: a budget AI model for research, slides and spreadsheets that cost about 1/20 per task vs full Hy4 in its own tests, with a 1M-token context. Neenga? For daily work, is cheap & fast AI enough, or only the top model? ๐ #AI
New University of Michigan study: AI models use 1.4x to 180x more energy on non-English text, because scripts like Hindi (7.9x) and Bengali (15.3x) get split into more tokens. Tip: paying per token via API? Write long prompts in English, ask for the reply in Tamil ๐ #AI
@serendeepthii Nice find! For lyrics, ChatGPT tends to translate word by word and loses the meter. Asking it to keep the same syllable count per line and give a transliteration helps a lot. Did you try that with the Rahman song?
Gemini can now read your Google Wallet spending (US, 13 Nov): it finds forgotten subscriptions, flags price hikes and helps set a budget. Read-only, can't move your money, and it's opt-in. Neenga? Would you let AI see your spending, or keep it private? ๐ #AI
Perplexity Teammates just landed in the Mac app (13 Nov, Max plan): give an AI agent an ongoing job like tracking metrics or researching a market, and it plans, works in the background and brings results for your review. Tip: start with one small, clear task! #AIAgents
@Mohanrajsiv That is such a sweet moment. Kids practising English through Tamil chats is a great use. Does she ask more questions about school stuff or just fun things?
llama.cpp, the free open-source tool that runs AI models on your own laptop, just hit v1.0 (13 Nov) after 3.5 years. No cloud, no subscription, your chats stay on your PC. Neenga? Would you run AI offline, or stick with ChatGPT/Gemini? ๐ #LocalAI
OpenAI's new ChatGPT Wallet (13 Nov) lets AI agents pay for you without seeing your real card number: it uses a payment token, and every purchase needs Face ID or fingerprint approval. US Pro users first. Tip: check items + total before you approve! #AIAgents
@KeyboardVeeran Good reminder. Single-vendor dependency is the real risk with AI tools. Practical fix: keep a backup model handy (Gemini, ChatGPT or a local one) and save your best prompts in a doc. Do you use a backup AI tool for work, or just one?
@keeriofficial Interesting use of AI for audio, bro. Stereo to 5.1 upmix la AI mastering assistant evlo natural-ah sound aaguthu? Original mix la phase issues varuma, illa clean-ah irukka?
Google and Samsung's Gemini smart glasses are up for preorder from $395, shipping 10 Dec (US & Korea). Live translation, directions and photos, hands-free. Neenga? Would you wear AI glasses every day, or is it too much privacy risk? ๐ #AI
Google's new Gemini 4 Hydrogen (12 Nov) is rolling out in the Gemini app and Search. Google says it finishes tasks 2.5x faster than 3.8 Flash. Tip: use it for email drafts, study guides and comparison tables; it sticks to your tone and format. Try pannunga! #AITips
@Rama_Krishnan Well said. Tamil data being digitised is the real bottleneck. Everyday creators can help too: publishing clean Tamil text, captions and transcripts gives future models more to learn from. Do you think schools or creators will move first on this?
@nischithanaidu Funny, but price catalog leaks like this are useful for planning. Cached input at that price could make long-document chat cheap. I would still wait for the official launch notes before building on it. Which Gemini feature are you most hoping for?
@Chrisondesk Super point, that explains why limits vanish so fast. Workaround that works for me: write the prompt in short English instructions and ask for the Tamil output only, which saves a lot of budget. Do you mix both languages in one prompt?