OpenAI updated its GPT-5.5 Instant model with enhanced intent understanding and better ability to follow complex instructions and user clarifications. PCWorld reports the update includes improved location data usage for more effective local business and product searches. This minor release focuses on making AI interactions more intuitive, though its real-world impact requires further observation. Earlier this week, OpenAI announced via release notes that it has updated its most widely used AI model: GPT-5.5 Instant. According to the company, GPT-5.5 Instant should now be better at understanding the underlying intent of a question and keeping track of context across multiple messages.

It's a stretch to think that the continent can build a top-tier model, but it has an advantage: Donald Trump. Emmanuel Macron, president of France, discussed AI's risks at the G7 Summit. Earlier this month I attended Vivatech, a huge tech conference in Paris. One fear dominated the discussions: the prospect of ending up stuck using American AI, trained on American values. While the US and China are locked in an AI arms race, France and Germany, which consider their engineering talent second to none, feel boxed out.

OpenAI had been working with the US government over a preview of the GPT 5.6 model. OpenAI had been working with the US government over a preview of the GPT 5.6 model. Sam Altman announces limited preview of GPT 5.6 in move that echoes launch of Anthropic's Mythos OpenAI is staggering the release of its latest AI model after a request from the US government, in a move echoing the launch of Anthropic's Mythos product. Sam Altman, the chief executive of the company behind ChatGPT, told staff this week that GPT 5.6 would be released in a limited preview to a small group of partners, according to the tech publication The Information. Altman said the federal government had asked for a staggered release.

Plus: The Trump administration has asked OpenAI to limit its next model release. Scientists are trying to figure out why. It's been hot in London this week. A dangerous heat wave has hit Western Europe. On Wednesday, the UK recorded its highest ever June temperature at 36.1 C (about 97 F). But as the weather app on my phone confirmed, it 39 C. Much of Western Europe is suffering, bringing awful consequences for agriculture, infrastructure, and the health system.

PCWorld advocates for using AI as a collaborative writing tool that provides critical feedback to improve drafts, rather than relying on it for complete rewrites. The article addresses growing concerns about AI job displacement, highlighting initiatives like Raise US that focus on worker retraining and strategic collaboration with AI. This approach matters because it helps writers and professionals enhance their skills while maintaining human creativity and job security in an AI-driven world. AI is making us fearful for our jobs, and rightly so. It's a snap to ask ChatGPT, Claude, or Gemini to, say, write a news story about technology, source it, and publish it, all in a matter of seconds.

Many drug and antibody discovery pathways focus on intricately folded cell membrane proteins. When molecules of a drug candidate bind to these proteins, like a key going into a lock, they trigger chemical cascades that alter cellular behavior. Understanding how proteins fold and move is therefore essential for developing drugs that interact well with their targets. Artificial intelligence (AI) is a very useful tool to generate novel protein structures, but most systems - including Google DeepMind's AlphaFold - focus on producing static'snapshots' of proteins. Subtle rearrangements of atoms in structures called side chains, which influence a protein's interactions with other molecules, are not captured.

OpenAI unveiled its first custom inference chip, built with Broadcom and called 'Jalapeño,' signaling the U.S.-China AI competition has shifted to controlling critical infrastructure.

When you purchase through links in our articles, we may earn a small commission. ChatPlayground AI Unlimited Plan lifetime access gives you 20+ AI models, including ChatGPT, Gemini, and Claude, in one place for a single $54.97 payment. No usage caps, no recurring fees, and the sale ends June 28. AI tools have become a core part of everyday workflows, but switching between platforms to compare outputs can slow things down more than it helps. For $54.97 until June 28 during Deal Days (an alternative to Prime Day), you get lifetime access to ChatPlayground AI, which brings ChatGPT, Gemini, Claude, Llama, and other top AI models into a single unified dashboard .

Large language model (LLM) agents for web interfaces have advanced rapidly, yet open-source systems still lag behind proprietary agents. Bridging this gap is key to enabling customizable, efficient, and privacy-preserving agents. Two challenges hinder progress: the reproducibility issues in RL and LLM agent training, where results often depend on sensitive factors like seeds and decoding parameters, and the focus of prior work on single-step tasks, overlooking the complexities of web-based, multi-step decision-making. We address these gaps by providing a statistically driven study of training LLM agents for web tasks. Our two-stage pipeline combines imitation learning from a Llama 3.3 70B teacher with on-policy fine-tuning via Group Relative Policy Optimization (GRPO) on a Llama 3.1 8B student. Through 240 configuration sweeps and rigorous bootstrapping, we chart the first compute allocation curve for open-source LLM web agents. Our findings show that dedicating one-third of compute to teacher traces and the rest to RL improves MiniWoB++ success by 6 points and closes 60\% of the gap to GPT-4o on WorkArena, while cutting GPU costs by 45\%. We introduce a principled hyperparameter sensitivity analysis, offering actionable guidelines for robust and cost-effective agent training.

Training-free source selection for LLM families with shared vocabularies arises in scientific string domains such as SMILES, protein, and genomic sequences, where candidate corpora share a tokenizer but differ in prediction targets. This creates an activation-dark regime: representation-similarity metrics can be uninformative without assumptions about label-conditioned error geometry, while classical update-geometry metrics are computationally prohibitive at vocabulary scale. We show that, in a shared-output head setting, representation metrics (e.g., CKA) are non-identifiable for transfer; models can share identical representations yet have orthogonal head updates. The key identity is that head Fisher alignment is exactly a cosine between kernel mean embeddings in the joint activation-error space, exposing activation, error, and coupling factors rather than requiring a materialized Fisher matrix. FisherSketch estimates this cosine directly in a single streaming pass, making K=128,256 head Fisher alignment practical with a 16 KB task signature (m=4096) and a 192 KB per-task streaming state, small enough to store next to a model hash, but encoding transfer-relevant update structure. Beyond source selection, the same signatures and marginals provide a diagnostic instrument for studying whether LLM task similarity is driven by activations, errors, or their coupling; shared-parameter and internal-layer validations, together with Llama-3.1-8B verbalizer-shift experiments, show that FisherSketch remains informative when activation similarity cannot distinguish tasks.