Jackson Atkins (@JacksonAtkinsX) / X

Jackson Atkins

2,215 posts

Jackson Atkins

@JacksonAtkinsX

Director of Engineering. Shipped million dollar systems in days. Surfacing AI breakthroughs in academic papers. Follow for AI intel before it hits mainstream.

United States

Joined April 2025

Replies Articles Media

Pinned
Jackson Atkins
@JacksonAtkinsX
Oct 7, 2025
My brain broke when I read this paper. A tiny 7 Million parameter model just beat DeepSeek-R1, Gemini 2.5 pro, and o3-mini at reasoning on both ARG-AGI 1 and ARC-AGI 2. It's called Tiny Recursive Model (TRM) from Samsung. How can a model 10,000x smaller be smarter? Here's how
Jackson Atkins
@JacksonAtkinsX
Aug 25, 2025
NVIDIA research just made LLMs 53x faster. 🤯 Imagine slashing your AI inference budget by 98%. This breakthrough doesn't require training a new model from scratch; it upgrades your existing ones for hyper-speed while matching or beating SOTA accuracy. Here's how it works:
Jackson Atkins
@JacksonAtkinsX
Jul 28, 2025
LLMs can now self-optimize. A new method allows an AI to rewrite its own prompts to achieve up to 35x greater efficiency, outperforming both Reinforcement Learning and Fine-Tuning for complex reasoning. UC Berkeley, Stanford, and Databricks introduce a new method called GEPA
Jackson Atkins
@JacksonAtkinsX
Aug 16, 2025
GPT-5 was a cost play. Make a somewhat better model at a fraction of the price. ARC AGI shows it. GPT-5 is about 10% better than o3-pro but costs 90% less.
Jackson Atkins
@JacksonAtkinsX
Oct 6, 2025
Everyone thinks Apple is sleeping on AI. Meanwhile Apple is building a massive graph knowledge base with 10s of millions of linked facts. You can't get data like this from a Google search. Apple is cooking. You just don't see it.
Jackson Atkins
@JacksonAtkinsX
Sep 6, 2025
Meta Superintelligence Labs just made LLMs handle 16x more context and unlocked up to a 31x speedup. 🤯 Their new REFRAG framework rethinks RAG from the ground up to achieve this, all with zero drop in accuracy. Here's how it works: The core problem with long context is
Jackson Atkins
@JacksonAtkinsX
Sep 1, 2025
Apple and Oxford just made AI 6.5x better at problem-solving. The secret: it teaches AI agents to ask perfect questions. This rockets success rates from 14% to 91%. No need for fine-tuning or retraining. It runs on current models. Here's how it works: It's a strategic loop
Jackson Atkins
@JacksonAtkinsX
Aug 2, 2025
AI just unlocked 3x more power from GPUs. A new AI framework called CUDA-L1 just taught itself to improve 250 different GPU tasks, delivering a 3.12x average speedup and a 120x peak gain. Here's how it works: The system's core is "Contrastive Reinforcement Learning
Jackson Atkins
@JacksonAtkinsX
Jul 21, 2025
Apple research just revealed a way to make LLMs 5.35x faster. 🤯 That’s not a typo. They've found a method to get a >500% speedup for code & math tasks, with ZERO quality loss. Here's how they're unlocking AI model's "latent potential": 🧵
Jackson Atkins
@JacksonAtkinsX
Oct 9, 2025
I woke up today to another 100x improvement in AI. 👀 Salesforce released a new AI training pipeline that gets the same results with 99% less data. It's called Webscale-RL and it uses 100x less tokens than continual pre-training. Here's how it works: The Webscale-RL pipeline
Jackson Atkins
@JacksonAtkinsX
Oct 7, 2025
Replying to @JacksonAtkinsX
The new AI model architecture Tiny Recursive Model (TRM) beats DeepSeek R1, Gemini 2.5 Pro, and o3-mini on ARC-AGI 1 and ARC-AGI-2. It has only 7 Million parameters and used 1000 training samples.
Jackson Atkins
@JacksonAtkinsX
Oct 25, 2025
Turns out AI can already do 65% of professional tasks... Nvidia had 38 PhDs and MBAs spend 10+ hours each creating benchmarks. They tested 40+ models on actual work. Not academic exercises. Real work that junior analysts and researchers do daily. GPT-5 scored 65.9% overall.
Jackson Atkins
@JacksonAtkinsX
Sep 21, 2025
MIT and Microsoft just made AI 64x better at planning, achieving 94% accuracy. 💥 Their PDDL-INSTRUCT method delivers a 66% absolute gain, teaching LLMs symbolic reasoning to validate their thoughts. This could be the next thinking milestone. Here's how it works: Educate on
Jackson Atkins
@JacksonAtkinsX
Sep 14, 2025
Meta just made training AI agents 25x faster. This is a breakthrough for robotics and complex planning. Meta's FAIR open sourced a new method called Scalable Option Learning. It trains a specialized agent at the scale previously seen only with LLMs. Here's how it works: The