Anjana

1.6K posts

Anjana banner
Anjana

Anjana

@anjanab_

IT professional Tracking AI model releases, benchmarks, AI tools, research breakthroughs & the future of intelligent

Pune Katılım Ekim 2023
137 Takip Edilen50 Takipçiler
Anjana
Anjana@anjanab_·
These papers are the foundation only. If you don't understand the Attention mechanism or how LoRA actually reduces trainable parameters, you'll struggle in any serious AI engineering role. Most people just import libraries without knowing the math.
Aditya@Aditya_181105

Research papers you must read for AI Engineer interviews: 1. Attention is all you need (Transformers) 2. LoRA (Low rank adaption) 3. PEFT ( Parameter Efficient Fine Tuning) 4. VIT (Vision Transformers) 5. VAE (Variational Auto Encoder) 6. GANs ( Generative Adversarial Networks) 7. BERT ( Bidirectional Encoder Representation from Transformers) 8. Diffusion Models (Stable Diffusion) 9. RAG (Retrieval Augment Generation) 10. GPT (Generative Pre-trained Transformers)

English
0
0
0
3
Anjana
Anjana@anjanab_·
RevPDF 4.0 just dropped with OCR and batch operations in only 20MB. This is how you build software. Adobe keeps adding heavy AI features that nobody asked for. Most people just want to edit a document offline without a subscription only. Small is better.
How To AI@HowToAI_

Someone built a free PDF editor that does everything Adobe does, in 20MB. RevPDF runs entirely on your device. Edit text and images, sign, redact, compress, convert to Word.. no internet, no account, no subscription. 100% Free. 100% Offline.

English
0
0
1
4
Anjana retweetledi
Reethu
Reethu@ritu_twts·
Be honest devs, Is coding still worth learning in the AI era?
Reethu tweet media
English
491
41
1.2K
239.7K
Anjana
Anjana@anjanab_·
Anthropic just made Claude available for Excel and Word, and Outlook is in beta. Some say this will finally automate the boring office work we all hate. Will this integration actually make you more productive, or is it just another AI gimmick?
English
0
0
0
9
Anjana
Anjana@anjanab_·
We are moving to a world where the model name is just the starting point. The provider's infra is the actual product. If you are not multi-cloud for your LLM inference yet, you are likely overpaying for slower tokens.
English
0
0
1
5
Anjana
Anjana@anjanab_·
The study found that by not sticking to one provider, you can optimize for what you actually need. DeepSeek-V3.2 saw a 90% speed increase through clever routing. This is basically free performance sitting on the table.
English
1
0
1
11
Anjana
Anjana@anjanab_·
Stop assuming all LLM APIs are created equal. A measurement study of hosted open-weight models just proved that the same model behaves differently depending on where you host it. The throughput gaps are massive. 🧵👇
English
1
0
1
12
Anjana
Anjana@anjanab_·
Anthropic and OpenAI just raised over 5 billion dollars combined specifically for enterprise services and custom deployments. With this much money going into services, will custom built AI setups beat standard off the shelf models for big companies?
English
0
0
0
6
Anjana
Anjana@anjanab_·
Most people think AI agents are just for small repetitive tasks. But the real shift is using them as an orchestration layer to connect all your company data and apps. type. venturebeat.com/ai/scaling-ai-…
English
0
0
1
7
Anjana
Anjana@anjanab_·
I have been tracking SGLang for a while now. It is a very solid framework for serving LLMs and multimodal models with high throughput. If you are struggling with inference latency or complex prompting workflows, this repo is worth only. github.com/sgl-project/sg…
English
0
0
1
9
Anjana
Anjana@anjanab_·
Centaur AI just dropped with a 40% jump in performance. This is huge for machine learning apps. - Better efficiency - Faster processing - Lower compute costs I am waiting to see if they can prove these numbers on actual production workloads.
English
0
0
0
4
Anjana
Anjana@anjanab_·
@morganlinton It is interesting how Codex and GPT-5.5 are winning on price and performance together. Anthropic is building a great enterprise ecosystem but their per-seat cost is getting very high for smaller teams. For a 16 person team, Cursor with GPT-5.5 is the more logical choice only.
English
0
0
1
29
Morgan
Morgan@morganlinton·
Officially canceling our Anthropic plan, it’s Codex + Cursor for my little 16 person eng team. Anthropic is great for companies that can spend $2,000/mo and up per engineer, but not affordable for us. Codex really upped their game recently, and with GPT 5.5, it’s just so good, and so token efficient. Still using Cursor plenty, my team still looks and reviews a lot of code. But with Cursor, we’ve never hit a limit, and Composer 2 is pretty awesome for most stuff. Testing out Droid as well and see some good early results with Droid + GLM 5.1, but still more testing to do before rolling it out to the whole team. My guess is many more engineering leaders will be sending messages like this. Anthropic makes great stuff but phew, it’s so darn token hungry. My team loves Codex and Cursor, onward!
Morgan tweet media
English
291
109
3K
336.2K
Anjana
Anjana@anjanab_·
@rezoundous OpenAI really fixed the hallucination issues in GPT-5.5 Instant, but Opus 4.7 is still better for heavy coding tasks. I feel GPT is more "human" now while Opus feels like a very strict textbook. Are you seeing GPT-5.5 handle multi-step reasoning better than the Opus updates?
English
0
0
0
168
Tyler
Tyler@rezoundous·
Can't explain it, but I trust GPT-5.5 more than Opus 4.7 right now.
English
329
70
3.5K
698.3K
Anjana retweetledi
ThePrimeagen
ThePrimeagen@ThePrimeagen·
without googling, i still cannot figure out how to copy and paste on a mac into: 1. the provided Terminal app (which sucks) 2. Ghostty. C-V does nothing, C-v goes into escape insertion (expected). I refuse to google this and it should just be obvious...
English
400
7
1.6K
325.3K
Anjana
Anjana@anjanab_·
Garry Tan is right about coding late. Most AI agents today fail because the prompts are too simple for complex logic. Meta-meta-prompting is basically building a brain for the agent only. If this scales, the $10B agent market might actually happen.
Garry Tan@garrytan

x.com/i/article/2052…

English
0
0
0
9
Anjana
Anjana@anjanab_·
Cleaning up PDFs for RAG is usually a nightmare but MinerU is actually quite good at it. It uses a dual VLM and OCR engine to handle 109 languages and turns messy docs into clean markdown. Definitely worth checking out if you are flows. github.com/opendatalab/Mi…
English
0
0
0
7
Anjana
Anjana@anjanab_·
Been checking out openclaw on GitHub. It is a personal AI assistant that runs locally and connects to 50 plus apps. The fact that it can write its own skills and run shell commands on your own hardware is huge for privacy. github.com/openclaw/openc…
English
1
0
4
19

Keşfet

@morganlinton @rezoundous @elonmusk @BarackObama @taylorswift13 @cristiano @BillGates @NASA