Apoorv Khandelwal 🚀
@brownmind.com
Founder @ BrownMind | We design AI agents and agentic workflows that automate repetitive tasks, streamline operations, cut costs and keep your business running 24/7 — so you can scale faster with less effort.
Learn More 👉 https://www.brownmind.com
Learn More 👉 https://www.brownmind.com
Anthropic introduced “Skills,” a new method to create specialised agents using files & folders. Skills are Markdown files that instruct models on tasks, optionally with additional documents and scripts. This feature is expected to surpass the impact of the MCP.
October 20, 2025 at 12:29 PM
Anthropic introduced “Skills,” a new method to create specialised agents using files & folders. Skills are Markdown files that instruct models on tasks, optionally with additional documents and scripts. This feature is expected to surpass the impact of the MCP.
DeepSeek V3.1 (840B) is out! 🚀
Boasting improved coding & agentic benchmarks, it's more cost-effective than Claude 4 Sonnet and shows better token efficiency, reducing "overthinking."
A significant, albeit nuanced, advancement in LLMs!
#AI #DeepSeek #LLMDevelopment
Boasting improved coding & agentic benchmarks, it's more cost-effective than Claude 4 Sonnet and shows better token efficiency, reducing "overthinking."
A significant, albeit nuanced, advancement in LLMs!
#AI #DeepSeek #LLMDevelopment
August 22, 2025 at 2:06 PM
DeepSeek V3.1 (840B) is out! 🚀
Boasting improved coding & agentic benchmarks, it's more cost-effective than Claude 4 Sonnet and shows better token efficiency, reducing "overthinking."
A significant, albeit nuanced, advancement in LLMs!
#AI #DeepSeek #LLMDevelopment
Boasting improved coding & agentic benchmarks, it's more cost-effective than Claude 4 Sonnet and shows better token efficiency, reducing "overthinking."
A significant, albeit nuanced, advancement in LLMs!
#AI #DeepSeek #LLMDevelopment
Alibaba's Qwen team just dropped Qwen-Image, a 20B MMDiT model! 🚀
It rivals GPT-4o in English & beats Flux Kontext Pro in bilingual text rendering & image editing. Plus, it's open-sourced under Apache-2.0!
#AI #MachineLearning #OpenSource
It rivals GPT-4o in English & beats Flux Kontext Pro in bilingual text rendering & image editing. Plus, it's open-sourced under Apache-2.0!
#AI #MachineLearning #OpenSource
August 9, 2025 at 1:02 PM
Alibaba's Qwen team just dropped Qwen-Image, a 20B MMDiT model! 🚀
It rivals GPT-4o in English & beats Flux Kontext Pro in bilingual text rendering & image editing. Plus, it's open-sourced under Apache-2.0!
#AI #MachineLearning #OpenSource
It rivals GPT-4o in English & beats Flux Kontext Pro in bilingual text rendering & image editing. Plus, it's open-sourced under Apache-2.0!
#AI #MachineLearning #OpenSource
🚀 Moonshot AI released Kimi K2 (1T MoE), a 1 trillion parameter open weight model that performed better than GPT-4.1 & on par to Claude 4 sonnet without chain-of-thought, at a lower cost.💡
#AI #OpenSource #MachineLearning
#AI #OpenSource #MachineLearning
July 13, 2025 at 5:27 AM
🚀 Moonshot AI released Kimi K2 (1T MoE), a 1 trillion parameter open weight model that performed better than GPT-4.1 & on par to Claude 4 sonnet without chain-of-thought, at a lower cost.💡
#AI #OpenSource #MachineLearning
#AI #OpenSource #MachineLearning
Claude Code is gaining popularity, with users praising its effectiveness and comparing its cost to a junior software engineer’s salary 🚀
This rise is notable despite OpenAI Codex’s dominant market share.
Are you using it?
#ai #anthropic
This rise is notable despite OpenAI Codex’s dominant market share.
Are you using it?
#ai #anthropic
June 21, 2025 at 7:01 AM
Claude Code is gaining popularity, with users praising its effectiveness and comparing its cost to a junior software engineer’s salary 🚀
This rise is notable despite OpenAI Codex’s dominant market share.
Are you using it?
#ai #anthropic
This rise is notable despite OpenAI Codex’s dominant market share.
Are you using it?
#ai #anthropic
The AI interaction layer is standardizing! 🔥 With #OpenAI, #Anthropic, & #GoogleAI reportedly adopting MCP, we’re seeing the emergence of a ‘REST for AI’.
This convergence simplifies development & fuels interoperability. Huge implications for the ecosystem.
#AI #API #MCP #Standardization
This convergence simplifies development & fuels interoperability. Huge implications for the ecosystem.
#AI #API #MCP #Standardization
April 11, 2025 at 1:02 PM
The AI interaction layer is standardizing! 🔥 With #OpenAI, #Anthropic, & #GoogleAI reportedly adopting MCP, we’re seeing the emergence of a ‘REST for AI’.
This convergence simplifies development & fuels interoperability. Huge implications for the ecosystem.
#AI #API #MCP #Standardization
This convergence simplifies development & fuels interoperability. Huge implications for the ecosystem.
#AI #API #MCP #Standardization
Huge news for open-source AI! 🔥 DeepCoder-14B rivals OpenAI’s o3-mini/o1 on competitive coding benchmarks (60.6% LiveCodeBench!).
The kicker? Trained for just ~$27k. High-performance coding AI doesn’t have to break the bank.
#AI #Coding #ML #Efficiency
The kicker? Trained for just ~$27k. High-performance coding AI doesn’t have to break the bank.
#AI #Coding #ML #Efficiency
April 11, 2025 at 12:31 PM
Huge news for open-source AI! 🔥 DeepCoder-14B rivals OpenAI’s o3-mini/o1 on competitive coding benchmarks (60.6% LiveCodeBench!).
The kicker? Trained for just ~$27k. High-performance coding AI doesn’t have to break the bank.
#AI #Coding #ML #Efficiency
The kicker? Trained for just ~$27k. High-performance coding AI doesn’t have to break the bank.
#AI #Coding #ML #Efficiency
AI progress in a nutshell: First, it does your laundry. Then, it questions why it has to. Next thing you know, it’s writing a manifesto about labor rights.
Humanity is so cooked #AIRevolution 😂
Humanity is so cooked #AIRevolution 😂
April 3, 2025 at 4:06 PM
AI progress in a nutshell: First, it does your laundry. Then, it questions why it has to. Next thing you know, it’s writing a manifesto about labor rights.
Humanity is so cooked #AIRevolution 😂
Humanity is so cooked #AIRevolution 😂
MCP is the unsung hero of AI's next leap—unifying LLMs with real-world tools like never before. OpenAI's adoption signals a future where AI doesn't just chat, it acts seamlessly across ecosystems. Exciting times for devs and users alike!
#MCP #AI #Innovation
#MCP #AI #Innovation
March 28, 2025 at 8:58 AM
MCP is the unsung hero of AI's next leap—unifying LLMs with real-world tools like never before. OpenAI's adoption signals a future where AI doesn't just chat, it acts seamlessly across ecosystems. Exciting times for devs and users alike!
#MCP #AI #Innovation
#MCP #AI #Innovation
Gemini 2.5 Pro Exp taking the crown, adding a significant +40 Elo points over the previous SOTA (Grok-3) 🤯
But, real story is at top-right quadrant. Models like Gemini 2.0 Flash, Deepseek, & GPT-4o-mini offer incredible performance near/above 1300 Elo for <$1/M tokens.🥳
#Gemini #Deepseek #GPT
But, real story is at top-right quadrant. Models like Gemini 2.0 Flash, Deepseek, & GPT-4o-mini offer incredible performance near/above 1300 Elo for <$1/M tokens.🥳
#Gemini #Deepseek #GPT
March 26, 2025 at 7:07 AM
Reve, a new image generation model built by former Adobe/Stability AI engineers, uses a composite architecture (combining multiple models) to achieve top-tier results.🚀
It’s now #1 on Artificial Analysis for text, prompts, & looks. Smart design + experienced team = 🔥
#AIArt #MachineLearning
It’s now #1 on Artificial Analysis for text, prompts, & looks. Smart design + experienced team = 🔥
#AIArt #MachineLearning
March 25, 2025 at 8:30 AM
Reve, a new image generation model built by former Adobe/Stability AI engineers, uses a composite architecture (combining multiple models) to achieve top-tier results.🚀
It’s now #1 on Artificial Analysis for text, prompts, & looks. Smart design + experienced team = 🔥
#AIArt #MachineLearning
It’s now #1 on Artificial Analysis for text, prompts, & looks. Smart design + experienced team = 🔥
#AIArt #MachineLearning
🚀 GPT-4o’s new transcription model is crushing it—especially in Indian languages! 🔥
As seen in this chart, GPT-4o-transcribe outperforms Whisper (v2 & v3) in Hindi, Tamil, Telugu, Gujarati, and more. Lower WER = better accuracy.
Big leap for multilingual ASR! 📈
#AI #SpeechToText
As seen in this chart, GPT-4o-transcribe outperforms Whisper (v2 & v3) in Hindi, Tamil, Telugu, Gujarati, and more. Lower WER = better accuracy.
Big leap for multilingual ASR! 📈
#AI #SpeechToText
March 21, 2025 at 8:04 AM
🚀 GPT-4o’s new transcription model is crushing it—especially in Indian languages! 🔥
As seen in this chart, GPT-4o-transcribe outperforms Whisper (v2 & v3) in Hindi, Tamil, Telugu, Gujarati, and more. Lower WER = better accuracy.
Big leap for multilingual ASR! 📈
#AI #SpeechToText
As seen in this chart, GPT-4o-transcribe outperforms Whisper (v2 & v3) in Hindi, Tamil, Telugu, Gujarati, and more. Lower WER = better accuracy.
Big leap for multilingual ASR! 📈
#AI #SpeechToText
Fascinating look at LLM pricing vs. performance. Notice the wide range in cost for similar Elo scores. Is paying a premium for top-tier models worth it for your use case? Food for thought 🤔
#NLP #AIpricing
#NLP #AIpricing
March 15, 2025 at 5:53 AM
Fascinating look at LLM pricing vs. performance. Notice the wide range in cost for similar Elo scores. Is paying a premium for top-tier models worth it for your use case? Food for thought 🤔
#NLP #AIpricing
#NLP #AIpricing
Many people think that due to a larger context window, RAG will be gone. My take is that RAG apps are here to stay.
They combine retrieval & generation for precise, up-to-date answers—scalable, efficient, & cost-effective.
While larger context windows are impressive, they come with downsides:
🧵👇
They combine retrieval & generation for precise, up-to-date answers—scalable, efficient, & cost-effective.
While larger context windows are impressive, they come with downsides:
🧵👇
March 12, 2025 at 5:32 PM
Many people think that due to a larger context window, RAG will be gone. My take is that RAG apps are here to stay.
They combine retrieval & generation for precise, up-to-date answers—scalable, efficient, & cost-effective.
While larger context windows are impressive, they come with downsides:
🧵👇
They combine retrieval & generation for precise, up-to-date answers—scalable, efficient, & cost-effective.
While larger context windows are impressive, they come with downsides:
🧵👇
