LLMs & Generative AI - Page 6 of 55
Latest breakthroughs in large language models and generative AI shaping the future of artificial intelligence and machine learning.
1086 articles
View complete article list
Anthropic receives US approval to relaunch Claude Mythos 5 model
Anthropic can sell its best brain again, with conditions. US regulators just approved a limited relaunch of the Claude Mythos 5 model. It’s the same restricted clearance OpenAI got for its GPT-5.6 Sol.
Routing Layer Cut AI Costs but Dropped Customer Satisfaction Scores
The math was brutal. A $100,000 monthly savings on inference costs, a tidy win for the engineering ledger. But the real numbers told a different story. Customer satisfaction cratered.
New Methods Let LLMs Auto‑Search Knowledge Bases, Replacing Manual Checks
Remember when you had to dig through a knowledge base yourself? That's over. The large language models are doing it now, and they aren't asking for permission.
GPT‑5.6 Sol outperforms GPT‑5.5 on GeneBench v1 genomics benchmarks
GPT-5.6 Sol is faster and cheaper. That’s the basic pitch. On the GeneBench v1 genomics test, it beats GPT-5.5 while using fewer tokens. The improvement is straightforward, which makes it notable.
ByteDance's iLLaDA Diffusion Model Generates Text 4× Faster, Scores Lower on MMLU
ByteDance built a language model that types four times faster than the usual kind. It also scores worse on tests. That’s the simple version. The model, called iLLaDA, is part of a quiet shift in AI.
AlgoEvolve uses LLMs to evolve and evaluate Python trading strategies
Most trading algorithms are built to follow rules. Then the rules break. A new research project, AlgoEvolve, tries something else. It gets large language models to write, test, and then continuously rewrite Python trading strategies until they work.
Company retains account, IP, session data despite “temporary” AI chats
The ghost icon promises invisibility. But the company still sees your account, your IP, your session. It hides the chat from your history, not from itself.
KRAFTON’s PUBG Ally uses NVIDIA ACE TTS and behavior trees for real‑time play
The shot rings out. You're pinned behind a wall, thirsting for ammo and a flank. Your human squadmate is down, but your other teammate, PUBG Ally, doesn't panic. It moves. It talks. It decides.
Physics‑Guided CNN Predicts Phase‑Separation Evolution in Binary Mixtures
Figuring out how a mixture of two liquids will separate over time is a classic physics problem. It’s also a massive computational headache.
OpenAI postpones GPT‑5.6 rollout after Trump administration request
The Trump administration asked. For once, Silicon Valley listened. OpenAI is holding back its next model, GPT-5.6. This isn't a delay for debugging.
Meta says AI moderators make 13% fewer errors than humans, defends rollout speed
Thirteen percent. Meta's entire case for automating content moderation hinges on that single statistic.
NVIDIA TensorRT Enables Context Parallelism for Multi‑GPU AI Inference
AI is hitting a wall with long prompts, and the transformer is to blame. Its attention mechanism has a quadratic scaling problem: double the sequence length, and processing demands quadruple. This quickly swamps the memory of any single GPU.
TokenSpeed-Kernel Delivers Top Performance on AMD GPT-OSS 120B via Gluon Kernels
LLM models and inference hardware are changing at breakneck pace. Why does that matter? Because speed alone isn’t enough any more.
OpenAI and Deepseek chatbots remain left‑leaning despite anti‑woke push
The numbers don’t lie, and they aren’t polite about it. A new investigation has put major AI chatbots to the test on political questions, and the results cut sharply against the marketing spin.
MiniCPM‑o 4.5 powers image understanding, captioning and text‑to‑image generation
Most vision AI can describe a scene, but its logic fails if you ask it to reason or create. Launched April 25, MiniCPM-o 4.5 tackles that trifecta directly.
Google adds screen-control to Gemini 3.5 Flash for cross‑platform agents
Google just taught an AI to use a mouse and keyboard. That's not a metaphor. Gemini 3.5 Flash can now look at your screen—on a phone, a browser, a desktop—and operate it. Click buttons. Type text.
LLM embeddings and HDBSCAN cluster text; visualized with pairwise scatterplots
Clustering text has always been a blunt job with blunt tools, a process that often butchers meaning just to fit a tidy spreadsheet. Machine Learning Mastery details a smarter path.
AI Agents Risk Fatal Traps When Treating Context Windows as Memory
Building an AI agent sounds like giving it a big brain. Often, it's just giving it a very long to-do list it can't finish. A quiet crisis is forming in agent design. The core mistake is simple but profound.
Two-Stage RAG Pipeline Uses Initial LLM Call to Match TOC Sections
Most retrieval-augmented generation is just expensive keyword search. It fails when the document gets too big. The problem is noise. Asking a model to find a specific clause in a 15,000-line contract is hopeless. The answer gets buried.
Harness-1 20B Model Beats GPT-5.4, Curates Top 8 Fairness‑Rated Results
Harness-1, a 20-billion-parameter AI subagent built for retrieval, now outperforms OpenAI’s GPT-5.4 model. The improvement came from a policy tweak.
Featured Resources & Reviews
No Code MBA Course Review
Learn to build AI-powered apps without coding. Our comprehensive review of No Code MBA's course.
AI Tools & Resources
Curated collection of AI tools, courses, and frameworks to accelerate your AI journey.
Weekly AI Digest
Get the week's most important AI news delivered to your inbox every week.