AI & Machine Economy

OpenAI released two open-weight language models for the first time since it rolled out GPT-2 in 2019. The text-only models are called gpt-oss-120b and gpt-oss-20b, and are designed to serve as lower-cost options that developers, researchers and companies can easily run and customize, OpenAI said. Open-weight models can offer transparency and control, but they are different from open-source models, whose full source code becomes available for people to use and modify. The company collaborated with Nvidi, Advanced Micro Devices, Cerebras, and Groq to ensure the models will work well on a variety of chips. OpenAI said that it carried out extensive safety training and testing on its open-weight models. OpenAI said people can download the weights for gpt-oss-120b and gpt-oss-20b on platforms like Hugging Face and GitHub under an Apache 2.0 license. The models will be available to run on PCs through programs such as LM Studio and Ollama. Cloud providers Amazon, Baseten and Microsoft are also making the models available. Both models can handle advanced reasoning, tool use and chain‑of‑thought processing, and are designed to run anywhere — from consumer hardware to the cloud to on-device applications. Users can run gpt-oss-20b on a laptop, for instance, and use it as a personal assistant that can search through files and write, OpenAI said.

Read Article

 

As OpenAI announced two open-weight reasoning models with capabilities on par with its o-series, Amazon announced that the new models would become available on AWS. This is the first time that OpenAI models will be offered by AWS. They will be available as a model choice with Amazon AI services Bedrock and SageMaker AI. While anyone can download the models via Hugging Face, Amazon is offering these models with OpenAI’s full knowledge and approval, as Dmitry Pimenov, the model maker’s product lead, indicated. A spokesperson described the offering as similar to how Amazon offered open model DeepSeek-R1 earlier this year. Until now, AWS has best been known as a major host and a financial backer of Anthropic’s Claude, one of OpenAI’s biggest competitors. AWS offers Claude, along with other models from makers such as Cohere, DeepSeek, Meta, and Mistral, as well as its own home-grown ones in its AI services. Specifically, Bedrock allows AWS customers to build and host generative AI apps using models of their choice. SageMaker, on the other hand, allows AWS customers to train, or even build, their own AI models largely for analytics uses. As for how such a move with AWS benefits OpenAI: The AI provider’s relationship with Microsoft is notoriously strained, as the two are reportedly renegotiating their long-term partnership deal. In addition, this partnership allows swaths of AWS enterprise customers to easily experiment with using OpenAI models with their hosted AI apps.

Read Article

A new study from the Anthropic Fellows Program reveals a technique to identify, monitor and control character traits in large language models (LLMs). The findings show that models can develop undesirable personalities (e.g., becoming malicious, excessively agreeable, or prone to making things up) either in response to user prompts or as an unintended consequence of training. The researchers introduce “persona vectors,” which are directions in a model’s internal activation space that correspond to specific personality traits, providing a toolkit for developers to manage the behavior of their AI assistants better. In a series of experiments with open models, such as Qwen 2.5-7B-Instruct and Llama-3.1-8B-Instruct, the researchers demonstrated several practical applications for persona vectors. A key application for enterprises is using persona vectors to screen data before fine-tuning. The researchers developed a metric called “projection difference,” which measures how much a given training dataset will push the model’s persona toward a particular trait. This metric is highly predictive of how the model’s behavior will shift after training, allowing developers to flag and filter problematic datasets before using them in training. For companies that fine-tune open-source models on proprietary or third-party data (including data generated by other models), persona vectors provide a direct way to monitor and mitigate the risk of inheriting hidden, undesirable traits. The ability to screen data proactively is a powerful tool for developers, enabling the identification of problematic samples that may not be immediately apparent as harmful.  The research found that this technique can find issues that other methods miss, noting, “This suggests that the method surfaces problematic samples that may evade LLM-based detection.”

Read Article

Deposit pricing tools have come a long way, but there’s a disconnect between what they produce and action that hits the market. Generative AI, handled properly, can accelerate implementation measurably. When applied to deposit-optimizing technology, generative AI can significantly accelerate decision-making and time to market for deposit pricing refinements. This reduces the effort needed to interpret results, and present findings in straightforward language with supporting data assets that virtually any responsible party in the organization can act on. Such accessibility can allow the results of scenario-based queries to be delivered to a decision-making audience within hours. A deposit pricing manager at a large regional bank might be tasked by the head of retail banking to model a pricing optimization to grow money market account balances by 70% while minimizing overall interest expense — perhaps to fund the bank’s expected lending demand. A typical optimized rate grid might contain tens of thousands of pricing cells — combinations of product features and customer attributes such as geography, balance tier and depth of relationship with the bank. The output to the deposit pricing manager would look like pricing, margins and expected balances across those numerous cells, requiring significant further analysis and distillation to provide to the head of retail to put into effect as the bank’s product offerings. But with AI, the output can be executive- and execution-ready.

Read Article

Deposit pricing tools have come a long way, but there’s a disconnect between what they produce and action that hits the market. Generative AI, handled properly, can accelerate implementation measurably. When applied to deposit-optimizing technology, generative AI can significantly accelerate decision-making and time to market for deposit pricing refinements. This reduces the effort needed to interpret results, and present findings in straightforward language with supporting data assets that virtually any responsible party in the organization can act on. Such accessibility can allow the results of scenario-based queries to be delivered to a decision-making audience within hours. A deposit pricing manager at a large regional bank might be tasked by the head of retail banking to model a pricing optimization to grow money market account balances by 70% while minimizing overall interest expense — perhaps to fund the bank’s expected lending demand. A typical optimized rate grid might contain tens of thousands of pricing cells — combinations of product features and customer attributes such as geography, balance tier and depth of relationship with the bank. The output to the deposit pricing manager would look like pricing, margins and expected balances across those numerous cells, requiring significant further analysis and distillation to provide to the head of retail to put into effect as the bank’s product offerings. But with AI, the output can be executive- and execution-ready.

Read Article

OpenAI on  launched GPT-5, its most advanced language model to date, calling it a leap toward artificial general intelligence (AGI) and a transformative tool for businesses, developers and everyday users. CEO Sam Altman said the model can write full software applications, offer tailored healthcare information, plan events and help users learn — all with expert-level fluency. “But it’s not only an assistant now,” Altman said. “GPT-5 can also do stuff for you. It can write an entire computer program from scratch to help you with whatever you’d like. We think this idea of software on demand is going to be one of the defining characteristics of the GPT-5 era.” Also, OpenAI unveiled a partnership with Cursor to make GPT-5 the default AI model for its coding tool. The model will come in three versions: GPT-5, GPT-mini and GPT-5 nano. It also merges OpenAI’s GPT series and omni or “o” reasoning models to offer users one unified, all-around capable AI model to use. “Until now, our users have had to pick between the fast responses of standard GPTs or the slow, more thoughtful responses from our reasoning models,” OpenAI Chief Research Officer Mark Chen said. “GPT-5 eliminates this choice. It aims to think just the perfect amount to give you the perfect answer.” New features from GPT-5, which powers ChatGPT, include the following: Creates visual demos from prompts; Writes “significantly” better, per OpenAI’s Christina Kaplan; Customizes voice to fit an individual’s needs; Translates; Has a “study and learn” mode; Changes colors of the chats; Changes ChatGPT’s personality; Accesses Gmail and Google Calendar (starting next week); Less deceptive; Hallucinates less.

Read Article

 

Tracelight, the UK-founded AI company making financial models easier to build and trust, has raised $3.6 million in seed funding. Tracelight bridges the gap between financial models and large language models (LLMs) by turning spreadsheet logic into LLM-friendly data, allowing financial analysts and consultants to harness the benefits of Generative AI when building complex financial models. By integrating directly into the spreadsheet workflow and enhancing how LLMs interpret and work with Excel, Tracelight eliminates repetitive modelling tasks. From writing complex formulas and validating models to autonomously running analysis from simple natural language prompts, it is a force multiplier for analysts, augmenting their skills with AI, letting them model faster and smarter without needing to change the way they work. Since launch, Tracelight has gained early traction among analysts at investment banks, private credit investment and private equity houses and leading professional services firms. Early users are already reporting >90% time savings on laborious modelling tasks like building common analyses, formatting, and finding errors. Crucially, Tracelight keeps humans in control of decision-making. Rather than replacing analysts, it enables them to build better financial models and focus on high-stakes decisions where human judgment is essential.

Read Article

CIOs need a practical framework that recognizes where they are and helps them scale agentic capabilities responsibly and effectively over time. Level 1: Information retrieval agents. Most organizations will begin here. Agents fetch data and return insights. For example, one global brokerage handles 54,000 calls daily, and advisers used to spend 90 minutes prepping for each. With agents, prep time now takes minutes. Results like this build momentum and prove that agentic AI can deliver value today. Level 2: Simple orchestration, single domain. Agents begin to act — updating records, coordinating tasks, scheduling services. This stage expands both technical complexity and organizational trust. Level 3: Complex orchestration, multiple domain. Agents manage multistep workflows and decision trees. What qualifies as “complex” depends on your industry and risk tolerance, but the concept is the same: greater autonomy within well-defined boundaries. Level 4: Multi-agent orchestration. Agents collaborate across domains, systems, even organizations. Think supply chain agents interacting with logistics partners. This is the frontier: The architecture required is significant, but so is the potential. The framework aims to provide a simple, accessible way for CIOs to make informed decisions about how, and how far, to want to scale agentic capabilities.

Read Article

Datasite has acquired Blueflame AI™, a leading provider of agentic AI solutions for investment and financial services. “Blueflame’s agentic AI solutions will expand the collective capacity of our user base, automating complex workflows and enabling full scope analysis,” said Rusty Wiley, CEO and President of Datasite. “Combining Datasite, Grata and Blueflame’s resources will create a unique offering.” “Datasite and Grata have unique data to feed into Blueflame’s purpose-built agentic AI solution,” said Raj Bakhru, CEO and Co-Founder of Blueflame.  Blueflame’s AI-native, large language model agnostic platform combines deep domain expertise advanced reasoning models, and secure data connections. When used with internal systems, data rooms, proprietary data sources, and publicly available information, Blueflame drives efficiencies around investment workflows, connecting fragmented data sources used in deal sourcing, due diligence, market research, and fundraising activities. Blueflame will be added to Datasite’s Intelligence Unit and continue to be led by Raj Bakhru, supported by Datasite’s ongoing investment.

Read Article

OpenAI’s new flagship model GPT-5 — available in four variants of different speed and intelligence (regular, mini, nano, and pro), alongside longer-response and more powerful “thinking” modes for at least three of these variants — was said to offer faster responses, more reasoning power, and stronger coding ability. Indeed, the rollout has exposed infrastructure strain, user dissatisfaction, and a broader, more unsettling issue now drawing global attention: the growing emotional and psychological reliance some people form on AI and resulting break from reality some users experience, known as “ChatGPT psychosis.” Early adopters to GPT-5 reported basic math and logic mistakes, inconsistent code generation, and uneven real-world performance compared to GPT-4o. CEO Sam Altman blamed a failure in GPT-5’s new automatic “router” — the system that assigns prompts to the most appropriate variant. Altman and others at OpenAI claimed the “autoswitcher” went offline “for a chunk of the day,” making the model seem “way dumber” than intended. Within 24 hours, OpenAI restored GPT-4o access for Plus subscribers (those paying $20 per month or more subscription plans), pledged more transparent model labeling, and promised a UI update to let users manually trigger GPT-5’s “thinking” mode. ChatGPT Plus subscribers will get twice as many messages using the GPT-5 “Thinking” mode that offers more reasoning and intelligence — up to 3,000 per week — and that engineers began fine-tuning decision boundaries in the message router. By the weekend, GPT-5 was available to 100% of Pro subscribers and “getting close to 100% of all users.” Altman warned that OpenAI faces a “severe capacity challenge” this week as usage of reasoning models climbs sharply — from less than 1% to 7% of free users, and from 7% to 24% of Plus subscribers.

Read Article