The Register
·
2 months ago
More than 3,000 software developers gathered at a conference organized by DeepLearning.AI in San Francisco to discuss how AI will reshape software development roles and practices. Speakers including executives from AMD and AWS emphasized reducing defect rates, spec-driven development, and the shift toward AI agents handling entire codebases rather than partial tasks. The consensus suggests software engineering will increasingly involve agent orchestration and management rather than manual coding, with development teams becoming smaller and more focused on overseeing AI-generated output.
The Register
·
2 months ago
● 3 sources
OpenAI's models are now available on Amazon Web Services' Bedrock platform, allowing enterprises to access OpenAI's GPT models without routing data through OpenAI's own APIs to address data privacy and security concerns. GPT-5.4 is available now with GPT-5.5 coming within two weeks, part of a deal involving up to $35 billion in financing contingent on OpenAI using two gigawatts of Amazon's Trainium accelerators. Microsoft renegotiated its OpenAI relationship to free OpenAI from revenue sharing commitments, enabling partnerships with other cloud providers while Microsoft remains the primary cloud provider.
The Register
·
2 months ago
AI researchers released Talkie, a 13-billion-parameter language model trained exclusively on English-language texts published before 1931 to study how AI systems behave with limited historical knowledge. The model underperforms compared to modern equivalents on standard benchmarks, achieving only 30 percent of performance when trained on OCR-transcribed pre-1931 texts versus human-transcribed versions. The researchers plan to scale Talkie to GPT-3.5 level capability by summer to explore scientific discovery, cultural change interpretation, and how language models form self-conception without knowledge of modern AI.
The Register
·
2 months ago
IBM announced general availability of Bob, an AI coding assistant that combines frontier LLMs, open-source models, and IBM's Granite SLM family to support the full software development lifecycle. Internal testing with 80,000 IBM employees reported 45 percent average productivity gains, with the Z mainframe package currently offered as a free technical preview and standard pricing ranging from $20 to $200 per month. The tool aims to help organizations analyze systems with technical debt and legacy documentation, though it has previously faced security vulnerabilities including malware manipulation and data exfiltration risks.
The Register
·
2 months ago
Amazon announced two AI services: an updated Amazon Quick desktop app that integrates with Google Workspace, Microsoft 365, Zoom, and Salesforce to automate tasks like meeting scheduling, and a significant overhaul of Amazon Connect into four agentic AI components for supply chains, hiring, healthcare, and customer experience. The Quick service was first unveiled in October 2025 as a web-based experience and now operates as a desktop application requiring only an email address to start. The moves position Amazon to compete with Microsoft Copilot and Salesforce Agentforce in the crowded market for AI agents that handle business processes, though existing customer relationships with other vendors present a significant adoption challenge.
Hugging Face Blog
·
2 months ago
● 2 sources
NVIDIA released Nemotron 3 Nano Omni, a multimodal AI model that processes text, images, video, and audio for document analysis, speech recognition, and computer control tasks. The model achieves 65.8 on OCRBenchV2 for document understanding and delivers 9x higher throughput compared to competing open models on multimodal workloads. Developers can now use a single model for processing long documents, transcribing speech, analyzing videos with narration, and automating graphical interface tasks.
The Register
·
2 months ago
Tenstorrent announced general availability of its Galaxy Blackhole AI compute platform, which packages 32 Blackhole accelerators into 6U systems offering 23 petaFLOPS of FP8 performance for $110,000 per node. A four-node Galaxy Supercluster costs $440,000 and can process a 100,000 token prompt in under four seconds while generating up to 300 tokens per second for DeepSeek V3. The improved software stack now supports considerably more models than before, with the company claiming 90 percent of Hugging Face models run on the platform, and Tenstorrent expects broader adoption through datacenter providers like Cirrascale and Equinix.
The Register
·
2 months ago
The European Commission has proposed measures requiring Google to allow competing AI services equal access to Android capabilities like sending emails and ordering food, which currently favor Google's own AI offerings. The Commission sent preliminary findings to Google under the Digital Markets Act in January and is holding public consultation on the proposals until May 13. Users would gain the ability to activate rival AI services via custom wake words and access deeply integrated experiences from multiple providers, increasing choice in how they interact with their phones.
The Register
·
2 months ago
● 2 sources
Enterprise organizations face mounting vendor lock-in as they discover switching between AI providers requires far more effort than anticipated, with 58 percent of migration attempts failing or requiring significantly more effort than expected. OpenAI raised GPT-5.2 pricing from $1.25 to $5.75 per input token, while Anthropic shifted Claude enterprise from fixed to usage-based pricing potentially doubling costs, and GitHub Copilot restricted compute availability and discontinued Opus model access. Companies increasingly dependent on specific vendor platforms, APIs, and integrated workflows will face sustained price increases as AI providers transition toward token-based pricing models reflecting real infrastructure costs.
The Register
·
2 months ago
India's four largest technology services companies—Infosys, HCL, TCS, and Wipro—reported mixed quarterly results, with all experiencing revenue pressure from what they term 'AI deflation,' where AI automation reduces the scope and cost of traditional IT services work. HCL expects annual revenue to decline 3-5 percent next year due to this deflation, while TCS reported 0.5 percent annual revenue decline and all four companies noted margin compression in some deals. The firms are responding by pivoting toward AI-related work such as digital twins, agentic AI solutions, and infrastructure modernization, though they have not implemented major layoffs despite the revenue headwinds.
The Register
·
2 months ago
● 2 sources
GitHub is switching Copilot from request-based to usage-based token billing on June 1, 2026, after subsidizing expensive inference costs became unsustainable. Copilot Pro subscribers will receive 1,000 AI Credits per month at $0.01 per credit, with complex models like Anthropic's Opus seeing pricing multipliers increase from 7.5x to 27x. Customers will face variable costs based on token consumption and can define overflow budgets or wait for monthly resets, while subscription rates remain unchanged but more expensive model usage is now explicitly metered.
Together AI
·
2 months ago
● 2 sources
Together AI made NVIDIA's Nemotron 3 Nano Omni multimodal model available on its inference platform, enabling developers to process video, images, audio, and text in a single reasoning pass. The 30B parameter model activates approximately 3B parameters per token and supports up to 256K tokens of shared context, reducing the need for fragmented multi-model pipelines. This simplifies deployment of agentic applications by eliminating separate inference passes across different modalities and lowering system complexity for production AI systems.
OpenAI Blog
·
2 months ago
OpenAI has implemented safeguards in ChatGPT including built-in model protections, detection systems for misuse, and policy enforcement mechanisms. The company works with external safety experts to identify and address potential harms. These measures aim to reduce risks from harmful outputs while maintaining the tool's utility for legitimate uses.
OpenAI Blog
·
2 months ago
● 2 sources
OpenAI's GPT models, Codex, and Managed Agents are now available through AWS, allowing enterprises to build AI applications within their own AWS infrastructure. The offering enables customers to access OpenAI's language and code generation capabilities without moving data outside their existing cloud environments. This integration reduces vendor lock-in concerns and provides enterprises with more control over their AI deployments and data governance.
