Back

Apple just exposed the truth behind so-called AI "reasoning" models like Claude, DeepSeek-R1, and o3-mini: They’re not actually reasoning — they’re just really good at memorizing patterns. Here’s what Apple found:

“reasoning models don't actually reason at all. they just memorize patterns really well” - apple

The underlying model of notebookLM has been upgraded to gemini 2.5 flash, but it is not 2.5 pro after all. The advantage is that the 2.5 series are reasoning models (thinking models), and the disadvantage is that flash is a distillation model.

Another open-source model has arrived, and it’s even better than DeepSeek-V3. The Allen Institute for AI just introduced Tülu 3 (405B) 🐫, a post-training model that is a fine-tune of Llama 3.1 405B, which outperforms DeepSeek V3.

🧠🍏 Apple Just Dropped a Bomb on AI Reasoning, “The Illusion of Thinking” Apple’s latest research paper is making waves and not the usual “faster chip, better cam” kind. In The Illusion of Thinking, Apple’s AI team basically said: “Yeah… large la

Most people I know are getting overly-reliant on AI for thinking/reasoning and strategy. This will have longterm negative impact on individuals as they'll become incapable of reasoning all together.

We’re so quick to trust anything that looks like thinking — especially in AI. But a recent paper (The Illusion of Thinking) shows this: Even the most advanced reasoning models collapse under real complexity. And strangely? They start thinking less,

Open ai has released the o3 pro model which is well enough to replace a senior software developer To make things worse it can be the foundational steps towards AGI by open ai First for the newbies we have two types of models Two types of models

AI is no longer just catching up — in many areas, it’s already ahead. According to the 2025 AI Index report, models now outperform humans in most technical tasks. From image classification to visual reasoning — machines are setting the pace. In j

India has no dearth of large language models (LLMs). Yet, most AI models struggle with the country itself 22 languages, multiple scripts, and patchy compute. Sarvam AI wants to fix this with its two indigenous models. But can it turn technical ambiti