f you work in fintech or finance, you already have too many tabs open and not enough time.

Fintech Takes is the free newsletter senior leaders actually read. Each week, I break down the trends, deals, and regulatory moves shaping the industry — and explain why they matter — in plain English.

No filler, no PR spin, and no “insights” you already saw on LinkedIn eight times this week. Just clear analysis and the occasional bad joke to make it go down easier.

Get context you can actually use. Subscribe free and see what’s coming before everyone else.

Sign Up Free

Telegram just rolled out a major AI-focused update that makes bots much more powerful inside the app. Users can now tag AI bots by username in private chats or group chats, even if the bot is not a member of the conversation. The bot only sees the message it was tagged in and replies to that specific thread, which gives users more control over privacy.

Telegram now lets bots respond to other bots, opening the door for more automated workflows and “agent-style” systems inside chats. Bots can also stream text as they generate it, so replies appear faster instead of arriving all at once.

Telegram also added chat automation for profiles, letting users connect a bot that can respond on their behalf in selected chats. Other updates include custom AI writing styles, search across more than 100 million emoji and stickers in 36 languages, poll statistics, country or subscriber-based poll limits, silent scheduled messages, and better reaction moderation for group admins.

Sony is partnering with Bandai Namco Holdings on a generative AI pilot focused on video production. Sony CEO Hiroki Totoki described AI as an “amplifier of human imagination,” while saying it should support creators rather than replace them. The companies say the pilot has already produced major speed and productivity gains, though Sony also admits consistency and control remain problems for professional creative work.

Sony Interactive Entertainment chief Hideaki Nishino also said AI could help speed up game development cycles and allow more creators to enter the market. This matters because big-budget games now take years to make, and studios are under pressure to produce more content faster.

Sony is already using AI in practical production tools. Studios like Naughty Dog and San Diego Studio have used Mockingbird, a facial animation tool that works after performance capture. Sony is also using AI to help with hair animation by feeding models videos of real hairstyles and generating detailed strand-level outputs.

The concern is volume. Nishino said AI could create a major increase in content, which raises the obvious fear: more low-quality “AI slop.” Sony says its studios will still focus on high-quality games, with human creators responsible for the vision, design, and emotional impact.

A new AI startup called Core Automation, led by former OpenAI researcher Jerry Tworek, is reportedly seeking funding at a $4 billion valuation just weeks after launch. The Information’s visible article summary says the company is part of a wave of new AI labs raising back-to-back rounds as investors chase startups founded by top AI talent.

Techmeme’s summary of The Information report says Core Automation was founded in March by Tworek and aims to raise $300 million to $500 million at that $4 billion valuation, after previously raising $100 million at a $1 billion valuation. Nvidia is reportedly among the earlier backers.

The startup is reportedly focused on building AI models that can keep learning after training, instead of staying mostly frozen until the next big retraining cycle. Business Insider also reported that Core Automation has attracted researchers from Anthropic and Google DeepMind and describes itself as “the world’s most automated AI lab.”

LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling

This paper introduces AutoTTS, a system that lets AI search for better ways to spend extra “thinking time” before answering. Instead of humans hand-building rules, it tests stored reasoning paths cheaply and finds strategies that improve accuracy for math tasks while keeping compute cost low. The full search cost only $39.90.

Listwise Policy Optimization: Group-based RLVR as Target-Projection on the LLM Response Simplex

This paper tries to make reward training—teaching AI by scoring its answers—more stable. It introduces Listwise Policy Optimization, which compares several answers at once, picks a clearer target, then moves the model toward it. In tests across logic, math, coding, and visual reasoning, it beat common training methods.

A^2RD: Agentic Autoregressive Diffusion for Long Video Consistency

This paper tackles a common AI video problem: long clips lose story and visual consistency. Its A²RD system makes video in segments, remembers earlier details, checks each new segment, and fixes mistakes before moving on. Across one-to-ten-minute tests, it improved consistency by up to 30% and story coherence by 20%.

OpenAI’s Chip Plan Hits a Microsoft Roadblock OpenAI’s custom chip deal with Broadcom has reportedly hit an $18 billion financing snag because Broadcom wants Microsoft to commit to buying 40% of the first chips produced. The chip, reportedly called “Jalapeno,” is meant to help OpenAI lower the huge cost of running AI models, but without Microsoft’s commitment, OpenAI may need to find new buyers or rework the deal.

Meta Is Building an AI Agent for Everyday Users Meta is reportedly building an AI helper called Hatch, inspired by OpenClaw, that could carry out tasks for regular users instead of only answering questions. The company is also working on an Instagram shopping tool that could help users find and buy products through AI before the fourth quarter.

Alibaba Brings Qwen AI Into Taobao Shopping Alibaba plans to connect its Qwen AI app with Taobao so shoppers can browse, compare, and buy products by chatting instead of typing search terms. The system will have access to more than 4 billion Taobao and Tmall products, plus features like delivery help, after-sales support, virtual try-ons, and 30-day price tracking.

Anthropic Eyes a Near-$1 Trillion Value Anthropic is reportedly considering a huge fundraising round that could raise up to $50 billion and value the company near $1 trillion. Investors are excited because revenue is rising fast, but the money is also needed to pay for massive computing power — the expensive chips and data centers needed to run advanced AI.