Back

Parampreet Singh

Python Developer ๐Ÿ’ป ...ย โ€ขย 9m

3B LLM outperforms 405B LLM ๐Ÿคฏ Similarly, a 7B LLM outperforms OpenAI o1 & DeepSeek-R1 ๐Ÿคฏ ๐Ÿคฏ LLM: llama 3 Datasets: MATH-500 & AIME-2024 This has done on research with compute optimal Test-Time Scaling (TTS). Recently, OpenAI o1 shows that Test-Time Scaling (TTS) can enhance the reasoning capabilities of LLMs by allocating additional computation at inference time, which improves LLM performance. In Simple terms: "think slowly with long Chain-of-Thought." But, By generating multiple outputs on a sample and picking the best one and training model again, which eventually leads to perform 0.5B LLM better than GPT-4o. But more computation. To make it efficient, they've used search based methods with the reward-aware Compute-optimal TTS. CC Paper: Can 1B LLM Surpass 405B LLM? Rethinking Compute-Optimal Test-Time Scaling ๐Ÿ”— https://arxiv.org/abs/2502.06703 #Openai #LLM #GPT #GPTo1 #deepseek #llama3

1 Reply
5
Replies (1)

More like this

Recommendations from Medial

Image Description

Chirotpal Das

Building an AI eco-s...ย โ€ขย 10m

I want a list of reasoning questions that openai o1 and/or deepseek r1 is failing to answer correctly. Quick help is much appreciated. Working on something and want to test it for reasoning capabilities.

1 Reply
5

Comet

#freelancerย โ€ขย 1y

Meet latest game-changer OpenAI o1 โ€“ an AI model built for deep thinking and unmatched performance - Takes extra time to think before responding for more accurate answers - Ranks in the 89th percentile in competitive coding (Codeforces) - Secure

See More
Reply
2
3

Swamy Gadila

Founder of Friday AIย โ€ขย 5m

๐Ÿšจ Open AI is an Wrapper๐Ÿ‘€๐Ÿคฏ Hot take, but letโ€™s break it down logically: OpenAI is not a full-stack AI company โ€” itโ€™s a high-level wrapper over Azure and NVIDIA. Hereโ€™s why that matters ๐Ÿ‘‡ ๐Ÿ”น 1. Infra Backbone = Microsoft Azure Almost 90%+ of Op

See More
Reply
2
4

Swamy Gadila

Founder of Friday AIย โ€ขย 1m

Big News: Friday AI โ€“ Adaptive API is Coming! Weโ€™re launching Adaptive API, the worldโ€™s first real-time context scaling framework for LLMs. Today, AI wastes massive tokens on static context โ€” chat, code, or docs all use the same window. The result?

See More
Reply
1
4
Image Description
Image Description

Vishu Bheda

ย โ€ขย 

Medialย โ€ขย 3m

๐—œ ๐˜€๐—ฝ๐—ฒ๐—ป๐˜ ๐Ÿฐ+ ๐—ต๐—ผ๐˜‚๐—ฟ๐˜€ ๐—ฟ๐—ฒ๐˜„๐—ฎ๐˜๐—ฐ๐—ต๐—ถ๐—ป๐—ด ๐—ž๐—ฎ๐—ฟ๐—ฝ๐—ฎ๐˜๐—ต๐˜†โ€™๐˜€ ๐—ฌ๐—– ๐—ธ๐—ฒ๐˜†๐—ป๐—ผ๐˜๐—ฒ. And I realized โ€” weโ€™ve been looking at LLMs the wrong way. Theyโ€™re not just โ€œAI models.โ€ Theyโ€™re a new kind of computer. โ€ข LLM = CPU โ€ข Context window = mem

See More
6 Replies
42
44

AI Engineer

AI Deep Explorer | f...ย โ€ขย 7m

LLM Post-Training: A Deep Dive into Reasoning LLMs This survey paper provides an in-depth examination of post-training methodologies in Large Language Models (LLMs) focusing on improving reasoning capabilities. While LLMs achieve strong performance

See More
Reply
2
Image Description
Image Description

Harsh Dwivedi

ย โ€ขย 

Medialย โ€ขย 4m

GPT-5 Full Review & 10 Mind-Blowing Use Cases OpenAI has just launched its most awaited model yet: GPT-5. And itโ€™s not just one step closer to AGI, but has almost entirely automated a lot of things using just simple prompts. In this video, we put t

See More
6 Replies
29
88
4
Image Description

Comet

#freelancerย โ€ขย 3m

GPT-5 is here.* This is the moment when AI stops being a shiny toy and becomes infrastructure. ๐Ÿ‘‡ OpenAI has launched its new flagship model, and the focus is clear: reliability, power, and a change in how we interact with AI. *The Essentials o

See More
1 Reply
2
11
Image Description
Image Description

Vishu Bheda

ย โ€ขย 

Medialย โ€ขย 1y

Jensen Huang, the CEO of NVIDIA, describes how AI is advancing in three key dimensions: 1. Pre-training: This is like getting a college degree. AI models are trained on massive datasets to develop broad, general knowledge about the world. 2. Post-

See More
5 Replies
2
13

Nupur Tevatiya

India's AI Filmmakin...ย โ€ขย 6m

๐Ÿšจ SmolVLA is here โ€” and itโ€™s changing how we think about robotics AI. Hugging Face just released SmolVLA, a lightweight Vision-Language-Action model trained on community-shared datasets from their LeRobot platform. Despite being just 450M paramete

See More
Reply
3

Download the medial app to read full posts, comements and news.