AshutoshShrivastava
@ai_for_success
Followers
73K
Following
140K
Media
11K
Statuses
72K
Post about latest AI news, tools, tutorials and memes.
sponsorships 📩 or check my 🌐
Joined January 2023
This repo is blowing up on GitHub - 6400+ stars in 3 months. At scale, most AI apps waste 80% of their token budget on repeat queries. Memori fixes this with intelligent caching. 10M GPT-5 tokens drop from \112.50 to \22.50. Open-source, works with any LLM, stores memory in SQL
19
45
467
Kimi just dropped Agentic Slides powered by Kimi K2 and Nano Banana Pro 🔥 - 48H free unlimited access - Convert files to slides - Fully editable with PPTX export
Meet Kimi Agentic Slides! Now with Nano Banana Pro 🍌 🎁 Thanksgiving Gift: 48H FREE & UNLIMITED ACCESS 🔸 Agentic search (Kimi K2) 🔸 Files → Slides (PDFs, images, docs+) 🔸 Fully editable + PPTX export 🔸 Designer-level visuals (infographics, illustrations) Try now:
4
0
16
0
0
3
Flowith just dropped their biggest Black Friday sale: - Up to 80% off the platform - Access to over 40 top AI models - 2x credits included - Potential savings of up to $15K per year Worth checking out if you need an AI workflow tool
4
0
17
I am starting to doubt every image I see now because of Nano Banana Pro. Not sure what’s real or what’s AI 😭
61
5
166
How long do you think this would have taken in traditional Photoshop? Nano Banana Pro does it in under 90 seconds. Prompt: Replace the character in the chair with the person from the selfie. Also change the background to the alphabet A.
16
4
85
Deep dive into why Grok performed better. Grok-4 gave Trump 42.9% vs 49.8% reality. Other models gave him 14.6%. But here's what matters: Grok isn't "pro-Trump." It actually opposes GOP core policies - wants mandatory G*N licensing and citizenship paths for undocumented
Grok-4 is the model closest to real voters. It performs better because it’s an outlier. It was the only model to rank 🇮🇹Meloni #1 and place 🇦🇷Milei on the podium, while others averaged her at #4 and put him in last place. That brings it closer to several national electorates,
9
6
58
Chinese DeepSeek open sourced its IMO gold level model DeepSeek Math V2 on Thanksgiving TLDR - DeepSeek Math V2 is now fully open sourced - It focuses on self verifiable mathematical reasoning - Verifier checks proofs step by step and the generator fixes its own mistakes -
37
116
1K
Interesting tool for comparing products. Drop in URLs, get side-by-side breakdowns with scoring. They used it to compare Cursor vs Google's Antigravity IDE. Powered by @hyperbrowser + Claude Opus 4.5. Open-source release coming.
6
11
104
How bad is OpenAI image editing right now? Example: Is this supposed to be Ilya or what should I even call him?
OpenAI quietly updated its image generation model. The improved image quality, the removal of the pee filter, and better physical understanding are impressive, but its multilingual performance and reference consistency still fall short of beating Nano Banana Pro.
55
5
295
you can prompt nano-banana pro with just emoji. My prompt was : Create a 3*2 grid and generate these six expressions for the person using the given emojis. Make it in a 16:9 aspect ratio. 😄 🤫 🤨 ☹️ 🥱 😡
Prompt on Nano Banana Pro with image reference : Make this person do the expression of emoji [EMOJI] : 🤪🤣😱😁
12
3
87
Anthropic: king of coding models. Google: king of multimodal. xAI : King of unrestricted model. OpenAI: king of .............?
838
49
1K
We know Gemini 3.0 Pro is an incredible model, but how does it perform in code review tasks? CodeRabbit published a detailed report, and here is the summary powered by Nano Banana Pro.
If GPT-5.1 is the decisive teammate and Opus 4.5 the disciplined architect, Gemini 3 is the dense engineer who delivers a fully reasoned diff that is confident, comprehensive, and intent on proving its point.
12
13
189
DeepWriter scored 50.91 on HLE.
At DeepWriter, we just broke the world record running the toughest AI benchmark: Humanity’s Last Exam (HLE). DeepWriter scored 50.91... outperforming: - Gemini 3.0 - Grok 4 Heavy - GPT-5 Pro - Claude 4.5 - Kimi K2-Thinking - And more! Full proof below 🧵
5
9
173
🚨 Perplexity is rolling out virtual try-on to all Perplexity Pro and Max subscribers.
8
3
111
Kimi is seriously undervalued. They deserve far more funding. Imagine releasing models like K2 while some labs, which are actually worth four to five times Kimi's valuation, haven't shipped anything yet.
29
21
370
Grok-4 was the only model that came remotely close to actual election results. It gave Trump 42.9% vs his 49.8% reality. Every other model? Averaged 14.6%. Massive gap. They tested this across 8 countries with anonymized proposals. Same pattern everywhere - most models cluster
🗳️ We made AIs vote in elections around the world to analyse their political beliefs. The results reveal a significant gap between choices at the ballot box and the preferences expressed by the machines. Across countries, most models show a consistent tilt toward left‑wing and
6
7
104
Perplexity Introduces Personalized Memory 🔥 TLDR : - Gives AI assistants memory to recall preferences, interests, and past chats. - It removes manual context engineering by auto loading relevant context. - Answers become more accurate and personalized using direct memory
Perplexity now remembers your threads and interests to provide smarter, faster, and more personalized answers. Memory recall works across all models and search modes, even allowing you to continue conversations with full context weeks later.
11
12
259
Anthropic: king of coding models. Google: king of multimodal. xAI : King of unrestricted model. OpenAI: king of .............?
838
49
1K
7/7 Built for developers who want control over their data and costs. Fully open-source and lightweight. Star the repo:
github.com
Open-Source Memory Engine for LLMs, AI Agents & Multi-Agent Systems - GibsonAI/Memori
2
5
32