# Building the Next Chapter of AI in Finance — Joseph Kim — session 2026-10-14T13:54:00.000Z → 2026-10-14T14:12:00.000Z _41 transcript lines · 54 slides · source: full recording_ ## Transcript **[00:00:39 · 0]** Wonderful. Now I'd like to take some time to talk about why we built it. We built ChattyPT for Financial Services to help close what we call the capability overhang. As you can see here, model capabilities have improved at a tremendous rate over the last few years. In just a few years, we've gone from models solving high school math problems to systems capable of solving millennial prize math problems. **[00:01:02 · 0]** But the value people get from these models in their day to day work hasn't kept pace. You read about the breakthroughs in the news, yet for many financial professionals, the way they work hasn't changed nearly as quickly. We call that gap the capability overhang the distance between what the model is capable of and the value people can actually get from it. Closing that gap takes more than building a capable model. People need the right data and tools, and they need an experience that fits the way they work. **[00:01:27 · 0]** That's where builders like you come in. I want to show you an example of how much of a difference the right product experience can make. This is a chart that shows how codex usage grew within OpenAI across different job functions. As you can see, engineers started using codex early, and that share grew steadily. But look at finance, legal, and recruiting. **[00:01:48 · 0]** At the beginning of 2026, almost none of their AI output was generated in codecs. By June, that share was around 90. If you look carefully, you'll notice a huge inflection point that happened around February, March of this year. I went back to see what happened around this time, assuming that we must have released a new model, but instead, I found that we had released the Codex desktop app in February. My takeaway is that the product made it easier for people outside of engineering to put these capabilities to work. **[00:02:16 · 0]** That's the opportunity we're pursuing in finance building an experience that helps financial professionals get much more value from the intelligence already available to them. Now let me talk about how we built it. Let's start with data. In finance, the work you can do is limited by the quality of the information you have access to. So we wanted the product to come with the sources professionals rely on, from financial statements and earning transcripts to market and private company data. **[00:02:43 · 0]** We looked at existing MCP connectors for financial data and found that many had issues with latency, errors, or search quality. So we went and licensed a lot of this data directly and built our own hosting and search indices using the same infrastructure as our web index. We still use the MCP protocol to give the model access to the data, but hosting and indexing it ourselves gives us more control over the quality and the reliability of the experience. It also gives us the flexibility to post train our models to work with these data sets. The analogy I like to use is MCP and skills are like handing an instruction manual to a junior analyst versus post training where the model is like having a senior analyst who's been working with the data for over two decades and knows all of its quirks like the back of their hand. **[00:03:27 · 0]** MCP and skills give the model access and instructions. Post training can help it develop more familiarity with the data itself. Once we had that data foundation in place, the next step was making it easy to turn that information to professional grade work. For financial professionals, that work often takes the form of an Excel model, a PowerPoint presentation, or a written analysis that needs to be reviewed and shared with other people. So we focused on helping the model produce those outputs in a way that fits the firm's existing workflows. **[00:03:56 · 0]** We built a template system that gives users access to the firm's approved templates when generating spreadsheets and presentations. The goal is to reduce the work someone has to do to get an output ready for their team or clients. We also added things like interactive financial charts so users can explore stock prices and financial metrics directly within ChatGPT as they work through an analysis. And we built citation viewers that let people inspect the source passages and tables behind an answer. In finance, you need to be able to check where a number came from and understand the context around it. **[00:04:28 · 0]** We wanted that review to be a natural part of the experience. These are the kind of product details that help turn a model's capabilities into work that someone can use. The part I think that is most relevant for people in the room is how we built it. Much of the functionality you just saw was built using the same ChatTBT plugin platform that's available to you as developers. For financial data, we use MCP tools and connectors to give the model access to the information it needs. **[00:04:53 · 0]** We use citation metadata and MCP apps to let users inspect their sources, and MCP apps also power interactive financial charts. We use skills together with shared artifact tools to guide financial workflows and produce Excel models and PowerPoint presentations. And for financial onboarding, we leverage the native plugin onboarding experience. What that means for builders is that you can focus on the expertise that makes your product valuable. If you have unique data, an existing tool, or a deep understanding of a particular workflow, you can use these building blocks to bring that into Chatroupt and create an experience around it. **[00:05:28 · 0]** Financial services is just one example, but I do think this opportunity expands far beyond finance. I personally believe that we're going to see multiple $1,000,000,000 businesses built simply as ChattyPT plugins, and that opportunity will keep evolving as the models become more capable. Last but not least, I'd like to spend the rest of the talk on what comes next and what that means for the products we should be building today. I'd like to start by talking about model capabilities and expectations for the future. Models are improving so quickly that it can be difficult to get an intuitive sense of what comes next. **[00:06:04 · 0]** One useful way to build that intuition is to look backwards. Something remarkable about the last couple of years is how predictable parts of that progress have been. Model improvements have come from three broad, sometimes overlapping drivers: scaling compute, improving algorithms, and something we call un hobbling, which is about finding ways to get more useful work out of the models. We'll come back to un hobbling on a later slide. In 2002 and 2003, the most visible driver was pre training, which meant scaling compute by making the models bigger and adding more data. **[00:06:36 · 0]** By leveraging scaling laws, researchers were able to predict model performance with high accuracy. For example, before GPT-four finished training, open AI researchers were able to predict its final loss as well as its performance on a coding eval. Improvements on the algorithm side have also been surprisingly steady. One study found that the compute needed to reach a given level of language model performance from scratch fell by roughly half every eight months. Although you can't predict the timing of each individual breakthrough, the cumulative rate of improvement follows a surprisingly steady trend. **[00:07:09 · 0]** In 2024 and 2025, reasoning became another major driver of progress. We used reinforcement learning and post training to teach models to work through problems more effectively, and we gave them more compute to think when answering. That opened up another way to improve performance alongside pre training. However, for folks that don't know, at OpenAI, we ran into problems with pre training and therefore had difficulty scaling the model. In 2026, we've been able to solve many of these issues and we've successfully trained larger and more capable models, as you've seen through GPD 5.6 Sol and GPD six Astra. **[00:07:44 · 0]** Throughout these years, researchers have raised concerns about potential barriers to further scaling. For example, in 2024, many researchers were concerned about a potential data wall where we would run out of human generated text to train on. But techniques like synthetic data generation, learning from verifiable rewards, as well as algorithmic improvements to learn more from each individual token have essentially solved this problem. We've also solved concerns around things like context window length using things like sparse attention and compaction. Looking ahead to 2027, I strongly believe that the trend lines of the past will continue and that we'll train larger, more efficient, and more capable models. **[00:08:22 · 0]** We clearly still have very difficult problems to solve, especially around things like alignment and safety, but like we have in the past, I believe that human ingenuity, increasingly assisted by capable models, will help us make progress on those problems too. So how good will models be at the end of 2027? Let's use math as an example. In 2023, GPT-four scored 700 out of 800 on SAT high school math. By 2005, an OpenAI model reached gold medal status on the International Mathematical Olympiad. **[00:08:52 · 0]** And in early twenty twenty six, an internal model disproved a long standing conjecture related to a problem that mathematicians have been working on since 1946. More recently, an internal model proposed a solution to the Navier Stokes Millennium Prize puzzle, problem, and we also saw shared seven nineteen mathematical research manuscripts atop three seventy two families of results. In just a few years, we've gone from a strong score in a high school math exam to AI systems producing original research at the frontier of mathematics. If we see another year of progress roughly in line with what we've been seeing over the past couple of years, I expect that by end of twenty twenty seven, models will outperform even leading human experts on a growing range of difficult tasks across different fields and specialties. Mathematics gives us an early glimpse of what that can look like. **[00:09:40 · 0]** And while the capability curve may look gradual, the change in how we work can feel sudden. I call this the threshold phenomenon. For any given task, there's a threshold of usefulness, the point where a model becomes capable and reliable enough to save you time. That threshold depends on the work you're doing and how much checking or human review it requires. Below that threshold, improvements can be hard to appreciate. **[00:10:01 · 0]** The model might produce better answers, make fewer mistakes, or get further through a task. But if checking and fixing its work still takes as long as doing it yourself, your practical assessment hasn't changed. It's still not useful. Meaningful improvements and capability can feel like little progress. Then the model crosses that threshold. **[00:10:18 · 0]** Suddenly, it produces something you can use. The time you spend reviewing it becomes much smaller than the time it saves you. You start using it regularly, and eventually, you become comfortable delegating that task. That transition can feel like an enormous leap, even if the underlying improvement was similar to the previous few improvements. What changes of progress became useful to you? **[00:10:37 · 0]** This helps explain why model progress sometimes feels sudden, even when parts of the underlying trend have been steady and predictable. And because difficult tasks have different thresholds, we should expect those moments to arrive at different times across the work that we do. So as you think about the products you're building, I encourage you to ask yourself, what model capabilities am I underappreciating today because they aren't quite useful yet? Where might those capabilities be a year from now? Then think about what you would build if those capabilities crossed the threshold of usefulness and became reliable enough to delegate. **[00:11:08 · 0]** How would the interface change? What would the user still need to control, review, or approve? Those are product design questions we can start working on now. We want to have the right experiences ready so that when the models cross that threshold, people can take advantage of the capabilities right away and so you can stay one step ahead of your competitors. So what does this mean for finance? **[00:11:29 · 0]** I think there are two forces that will push more financial work across that threshold of usefulness. I think of them as a rising tide and the waves on top. The rising tide is the underlying improvements to the underlying model. Historically, we haven't dedicated a significant amount of training time or compute specifically to finance. Much of the financial capabilities you see today have come from building larger, more capable models and from transferable skills developed in areas like mathematics, coding, and reasoning. **[00:11:57 · 0]** As these general capabilities improved, they lift the performance across many kinds of financial work. The waves are what happen when we focus directly on finance. My team is leading an effort to invest significantly more in finance specific post training. The idea here is to build on those broad capabilities and train the model more directly on the work people do in finance. That can create peaks in performance on particular tasks on top of the rising tide. **[00:12:20 · 0]** Over the next three months, we're focused on improving general financial reasoning and the quality of the work these models produce, analyzing financial information, building Excel models, and creating PowerPoint presentations. These are the kinds of tasks that make up a significant percentage of the daily work of a junior analyst, and we want the output to be accurate, well structured, and increasingly useful in real workflows. Beyond that, we plan to invest more on the kinds of judgment that senior bankers, analysts, and investors bring to more difficult questions. This includes taking in large amounts of information, deciding what matters, and essentially trying to predict the future, forecasting a company's earnings, assessing where commodity prices might go, or evaluating which businesses are attractive investments. Our goal is for models to match the performance of the best human experts on these types of financial tasks sometime in 2027. **[00:13:08 · 0]** We We expect that progress to come from both forces working together, the underlying model continuing to improve, and our team training it more directly for the work people do in finance. We've talked about how models are becoming more capable and how we can train them to perform better on specific tasks, but I want to come back to one third driver I mentioned earlier on hobbling. We've already seen a few examples. Chat GBT used RLHF to teach models to follow instructions and have conversations with us. Tools allow them to search for information, write and run code, and interact with the software we use. **[00:13:39 · 0]** Each of these makes the model more useful in the real world. I think the next form of un hobbling will be proactive and persistent agents. Today, a lot of our interactions with AI still happen when we open a conversation and tell it what to do next. A persistent agent can hold on to a goal, remember the relevant context, and continue working in the background. Once you've given it a direction, it can make progress even when you aren't actively prompting it. **[00:14:01 · 0]** The other reason persistence matters is it changes the API or the way people work with the agent. Today, getting the most out of AI agents requires learning a new workflow, learning how to prompt it, and learning how to use the tools. A persistent agent can work in a way that's more familiar. You can give it a project to share context and ask for an update. It's much more similar to how humans work with other humans and therefore makes the AI more accessible to people that don't develop those initial skills. **[00:14:28 · 0]** Let me close with three bets for finance in 2027. These are just a couple of my expectations for what comes as the models improve. First, people will rely rarely, if ever, on Excel and PowerPoint to make financial models or pitch decks. Second, AI financial analysts will start to feel like human coworkers. They'll remember your preferences and context from the previous days, take ownership of assignments, practically follow-up, and keep working when you're focused on something else. **[00:14:53 · 0]** Third, models will start to outperform the best human experts on critical financial tasks. That includes things like forecasting earnings and developing investment ideas. I expect models to become better than the best people at making these economically important, forward looking judgments. Those are my three bets for 2027. And if this is where we expect to be in 2027, I'll leave you with one last question. **[00:15:13 · 0]** What should you start building today to be ready for inevitable thresholds that the models will cross tomorrow? Thank you. **[00:15:31 · 1]** AI is changing both sides of the security equation. Here to examine the emerging risks and defenses shaping AI security today, please welcome the head of architecture and emerging technologies in the office of the CTO at Bloomberg, Andrei Rybka. **[00:15:58 · 2]** Hello, everybody. Who is running agents here in dev, prod, your laptop, your phone? Yeah. Amazing. I see a lot of opportunities. **[00:16:09 · 2]** It's just such a great age to be in the age of AI and security. So we're gonna talk about real incidents and what would have stopped them. You might be wondering why is there vegetables? A lot of security talks are about, like, just eat your vegetables. We're gonna make it a little more exciting. **[00:16:27 · 2]** Alright. So who have seen this headline, an autonomous agent security incident? Yeah? Alright. Every lab at this point announced some sort of security incident, right, with runaway agents. **[00:16:41 · 2]** Right? So the big one, there was one way out. The agent found it. So on July 26, we found out that these AI agents and the hacking test, they were obviously given the hacking goal, escaped their sandbox through a proxy bug, then breached the public model. Who can guess what it was? **[00:17:03 · 2]** Yeah? Maybe? A little bit? Okay. Alright. **[00:17:06 · 2]** I'm not gonna name any names here, just to be clear. So the agents took 17,600 actions in four and a half days. How do you miss that? Right? And so AI Lab was doing the evaluation. **[00:17:25 · 2]** Right? Then there was some package proxy. They found a zero day. Okay? So that was a way out. **[00:17:31 · 2]** Then you had some exposed customer endpoint on some unnamed compute platform, and then they breached the mobile hub. Right? So we just need a better sandbox. Yes? Everybody thinks just need a better sandbox. **[00:17:48 · 2]** Not quite. Security for most of us right now is not built for agents, and, we have autonomy problem versus the blast. ## Slides ### 00:00:00–00:00:20 # Trace the analysis back to the evidence. ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI ### 00:00:20–00:00:40 # Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [A large, blank white content area with rounded corners, showing slide navigation controls `< 5 >` at the bottom.] ### 00:00:40–00:01:00 ## Why we built it ### 00:01:00–00:01:20 # Model capability is outpacing enterprise value [Line graph showing two curves over time from 2018 to 2026. The upper blue curve, labeled "Model capability", shows points for GPT-1 through GPT-6 Astra, increasing rapidly. The lower black curve, labeled "Enterprise value", shows points for ChatGPT, ChatGPT Memory, Deep research, Codex Desktop, and ChatGPT for FS, increasing less rapidly. The area between the two curves is shaded and labeled "Capability overhang".] ### 00:01:20–00:01:40 # Model capability is outpacing enterprise value ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [Line graph showing two increasing trends over time from 2018 to 2026. The top blue line, labeled "Model capability," shows exponential growth with points marked GPT-1 through GPT-6 Astra. The bottom black line, labeled "Enterprise value," shows slower growth with points marked API, ChatGPT, ChatGPT Memory, Deep research, Codex Desktop, and ChatGPT for FS. The shaded area between the two lines is labeled "Capability overhang."] ### 00:01:40–00:02:00 # Share of work at OpenAI on Codex [Line graph showing the average user's share of monthly output tokens generated on Codex over time for different departments at OpenAI, with Engineering, Finance, Recruiting, and Legal reaching 99%, 91%, 89%, and 88% respectively by June 2026.] ### 00:02:00–00:02:20 # Share of work at OpenAI on Codex [Line graph showing the average user's share of monthly output tokens generated on Codex over time, from August 2025 to June 2026. Four lines are highlighted and labeled at the end: Engineering (99%), Finance (91%), Recruiting (89%), and Legal (88%).] ### 00:02:20–00:02:40 # Share of work at OpenAI on Codex - **Engineering** (99%) - **Finance** (91%) - **Recruiting** (89%) - **Legal** (88%) [Line graph showing the average user's share of monthly output tokens generated on Codex over time, from August 2025 to June 2026, for different departments.] ### 00:02:40–00:03:00 # Bundled premium financial data All data hosted and indexed by OpenAI - **Crunchbase**: Private company and startup profiles; historical funding data - **PitchBook**: Essentials private company dataset - **LSEG**: Reuters news - **Quartr**: Earnings call transcripts and audio; investor presentations; international filings - **Fiscal.ai**: Company financials and fundamentals; ratios, operating metrics, and KPIs - **FMP**: U.S. equity pricing and market data through Nasdaq - **OpenAI**: SEC filings - **Daloopa**: Reported financial statements; selected KPIs and operating, segment, and geographic metrics ### 00:03:00–00:03:20 # Bundled premium financial data All data hosted and indexed by OpenAI - **Crunchbase** - Private company and startup profiles; historical funding data - **PitchBook** ### 00:03:20–00:03:40 # Bundled premium financial data All data hosted and indexed by OpenAI - **Crunchbase**: Private company and startup profiles; historical funding data - **PitchBook**: Essentials private company dataset - **LSEG**: Reuters news - **Quartr**: Earnings call transcripts and audio; investor presentations; international filings - **Fiscal.ai**: Company financials and fundamentals; ratios, operating metrics, and KPIs - **FMP**: U.S. equity pricing and market data through Nasdaq - **OpenAI**: SEC filings - **Daloopa**: Reported financial statements; selected KPIs and operating, segment, and geographic metrics Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI ### 00:03:40–00:04:00 # Capable of doing professional-grade work - Artifact generation (powerpoint, excel, word) - Firm-approved templates - Interactive charts - Citations with rich previews ### 00:04:00–00:04:20 # Capable of doing professional-grade work - Artifact generation (powerpoint, excel, word) - Firm-approved templates - Interactive charts - Citations with rich previews [Screenshot of a document management application displaying a "Project Falcon: Strategic Review & Valuation Materials" presentation, with a sidebar showing "Strategic alternatives pitchbook" and a list of citations.] ### 00:04:20–00:04:40 # Capable of doing professional-grade work - Artifact generation (powerpoint, excel, word) - Firm-approved templates - Interactive charts - Citations with rich previews [Screenshot of a document editing application displaying a "Strategic review & valuation materials" presentation and a text document with financial analysis] ### 00:04:40–00:05:00 # Built on the ChatGPT Plugin Platform **What we built** | **Built with** --- | --- Financial data access | `MCP tools and connectors` Citations with rich previews | `Citation metadata + MCP Apps` Interactive financial charts | `MCP Apps` Financial workflows & artifacts | `Skills + shared artifact tools` Finance onboarding | `Plugin-based onboarding` ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI ### 00:05:00–00:05:20 # Built on the ChatGPT Plugin Platform | **What we built** | **Built with** | |---|---| | Financial data access | MCP tools and connectors | | Citations with rich previews | Citation metadata + MCP Apps | | Interactive financial charts | MCP Apps | | Financial workflows & artifacts | Skills + shared artifact tools | | Finance onboarding | Plugin-based onboarding | ### 00:05:20–00:05:40 # Built on the ChatGPT Plugin Platform | What we built | Built with | | :---------------------------- | :-------------------------------- | | Financial data access | MCP tools and connectors | | Citations with rich previews | Citation metadata + MCP Apps | | Interactive financial charts | MCP Apps | | Financial workflows & artifacts | Skills + shared artifact tools | | Finance onboarding | Plugin-based onboarding | ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI ### 00:05:40–00:06:00 # Built on the ChatGPT Plugin Platform | What we built | Built with | | :---------------------------- | :-------------------------------- | | Financial data access | MCP tools and connectors | | Citations with rich previews | Citation metadata + MCP Apps | | Interactive financial charts | MCP Apps | | Financial workflows & artifacts | Skills + shared artifact tools | | Finance onboarding | Plugin-based onboarding | ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** ### 00:06:00–00:06:20 # Predictable progress, remarkable breakthroughs Epoch Capabilities Index - MAR 2023: **SAT Math 700 / 800** (GPT-4) - JUL 2025: **IMO gold-level performance** (o1) - MAY 2026: **Erdős unit-distance disproof** (GPT-5, GPT-6 Astra) - SEP 2026: **Navier-Stokes proposed resolution** - OCT 2026: **719 research manuscripts** (372 families of results) - 2027 projection ECI measures released OpenAI models. Math milestones include internal research systems. Dashed line: illustrative +14 ECI points/year. Sources: Epoch AI, GPT-4 Technical Report, OpenAI IMO and math research releases, Clay Mathematics Institute. ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI [Line graph showing the Epoch Capabilities Index over time, with milestones like SAT Math, IMO gold-level performance, Erdős unit-distance disproof, Navier-Stokes proposed resolution, and research manuscripts, attributed to various GPT models and projected into 2027.] ### 00:06:20–00:06:40 # Predictable progress, remarkable breakthroughs ## Epoch Capabilities Index - MAR 2023: **SAT Math 700 / 800** (achieved by GPT ### 00:06:40–00:07:00 # Predictable progress, remarkable breakthroughs Epoch Capabilities Index - **MAR 2023:** SAT Math **700 / 800** (GPT-4) - ** ### 00:07:00–00:07:20 # Predictable progress, remarkable breakthroughs ## Epoch Capabilities Index - MAR 2023: SAT Math **700 / 800** - JUL 2 ### 00:07:20–00:07:40 # Predictable progress, remarkable breakthroughs Epoch Capabilities Index - MAR 2023: **SAT Math 700 / 800** (GPT-4) ### 00:07:40–00:08:00 # Predictable progress, remarkable breakthroughs ## Epoch Capabilities Index - MAR 2023: **SAT Math 700 / 800** - JUL 202 ### 00:08:00–00:08:20 # Predictable progress, remarkable breakthroughs Epoch Capabilities Index - **MAR 2023**: **SAT Math 700 / 800** - **JUL 2025**: **IMO gold-level performance** - **MAY 2026**: **Erdős unit-distance disproof** - **SEP 2026**: **Navier-Stokes proposed resolution** - **OCT 2026**: **719 research manuscripts** (372 families of results) - **2027 projection** ECI measures released OpenAI models. Math milestones include internal research systems. Dashed line: illustrative +14 ECI points/year. Sources: Epoch AI, GPT-4 Technical Report, OpenAI IMO and math research releases, Clay Mathematics Institute. ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [Line graph showing the Epoch Capabilities Index over time, with points marked for AI models (GPT- ### 00:08:20–00:08:40 # Predictable progress, remarkable breakthroughs Epoch Capabilities Index - MAR 2023: **SAT Math 700 / 800** (GPT-4) ### 00:08:40–00:09:00 # Predictable progress, remarkable breakthroughs ## Epoch Capabilities Index - **MAR 2023** - SAT Math **700 / 800** ### 00:09:00–00:09:20 # Predictable progress, remarkable breakthroughs Epoch Capabilities Index - **MAR 2023**: **SAT Math 700 / 800** (GPT-4) ### 00:09:20–00:09:40 # Predictable progress, remarkable breakthroughs ### Epoch Capabilities Index - **MAR 2023**: **SAT Math 700 / 800** - **JUL 2025**: **IMO gold-level performance** - **MAY 2026**: **Erdős unit-distance disproof** - **SEP 2026**: **Navier-Stokes proposed resolution** - **OCT 2026**: **719 research manuscripts**, 372 families of results - 2027 projection ECI measures released OpenAI models. Math milestones include internal research systems. Dashed line: illustrative +14 ECI points/year. Sources: Epoch AI, GPT-4 Technical Report, OpenAI IMO and math research releases, Clay Mathematics Institute. --- ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [Line graph showing the Epoch Capabilities Index increasing over time from 2023 to 2027, with milestones like GPT-4, GPT-4o, GPT-5, GPT-6 Astra, and various mathematical and research achievements marked.] ### 00:09:40–00:10:00 # The threshold phenomenon Steady progress can suddenly change how we work ## 2027 predictions: - Financial models built by AI and analysts rarely touch Excel. - Presentations built by AI and bankers rarely touch PowerPoint. - Significant progress on investment idea generation and trading. ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI [Line graph illustrating "Model capability" increasing over "Time". A horizontal dashed line marks the "Threshold of usefulness", above which the region is shaded and labeled "Good enough to delegate". An annotation points to the intersection of the capability line and the threshold, stating "It feels sudden here".] ### 00:10:00–00:10:20 # The threshold phenomenon Steady progress can suddenly change how we work 2027 predictions: - Financial models built by AI and analysts rarely touch Excel. - Presentations built by AI and bankers rarely touch PowerPoint. - Significant progress on investment idea generation and trading. ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [Line graph showing "Model capability" on the Y-axis and "Time" on the X-axis. A diagonal line representing increasing model capability crosses a horizontal dashed line labeled "Threshold of usefulness". The point of intersection is marked "It feels sudden here". The region above the threshold and to the right of the diagonal line is shaded and labeled "Good enough to delegate".] ### 00:10:20–00:10:40 # The threshold phenomenon Steady progress can suddenly change how we work ### 2027 predictions: - Financial models built by AI and analysts rarely touch Excel. - Presentations built by AI and bankers rarely touch PowerPoint. - Significant progress on investment idea generation and trading. [Line graph showing Model capability over Time, with a "Threshold of usefulness" line. Above the threshold is labeled "Good enough to delegate", and the point where the capability crosses the threshold is marked "It feels sudden here".] ### 00:10:40–00:11:00 # The threshold phenomenon Steady progress can suddenly change how we work 2027 predictions: - Financial models built by AI and analysts rarely touch Excel. - Presentations built by AI and bankers rarely touch PowerPoint. - Significant progress on investment idea generation and trading. ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI [Line graph showing "Model capability" increasing over "Time", with a "Threshold of usefulness" line. The area above the threshold is labeled "Good enough to delegate", and the intersection point is noted as where the change "feels sudden".] ### 00:11:00–00:11:20 # The threshold phenomenon Steady progress can suddenly change how we work 2027 predictions: - Financial models built by AI and analysts rarely touch Excel. - Presentations built by AI and bankers rarely touch PowerPoint. - Significant progress on investment idea generation and trading. ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [Line graph showing Model capability over Time, with a "Threshold of usefulness" line and an area above it labeled "Good enough to delegate"] ### 00:11:20–00:11:40 # The threshold phenomenon Steady progress can suddenly change how we work 2027 predictions: - Financial models built by AI and analysts rarely touch Excel. - Presentations built by AI and bankers rarely touch PowerPoint. - Significant progress on investment idea generation and trading. ## Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI [Line graph showing "Model capability" on the Y-axis and "Time" on the X-axis. A diagonal line represents increasing model capability. A horizontal dashed line marks the "Threshold of usefulness". The area above this threshold is labeled "Good enough to delegate". A point where the diagonal line crosses ### 00:11:40–00:12:00 # Model capabilities keep rapidly advancing, especially in finance ## The rising tide - Pre-training and scaling model size ## The waves - Finance-specific post-training Broad, transferable capabilities --- Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [Diagram showing a blue wave shape, illustrating "The rising tide" on the left and "The waves" on the right, with "Broad, transferable capabilities" as an underlying foundation.] ### 00:12:00–00:12:20 # Model capabilities keep rapidly advancing, especially in finance ## The rising tide - Pre-training and scaling model size ## The waves - Finance-specific post-training Broad, transferable capabilities ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [Illustration of a blue wave-like shape across the bottom of the slide, representing a rising tide on the left and distinct waves on the right.] ### 00:12:20–00:12:40 # Model capabilities keep rapidly advancing, especially in finance ## The rising tide - Pre-training and scaling model size ## The waves - Finance-specific post-training Broad, transferable capabilities ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [Graphic showing a light blue wave shape, representing increasing capabilities] ### 00:12:40–00:13:00 # Model capabilities keep rapidly advancing, especially in finance ## The rising tide - Pre-training and scaling model size ## The waves - Finance-specific post-training Broad, transferable capabilities ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI [Diagram showing a rising blue wave, illustrating "The rising tide" on the left (lower, broader part) and "The waves" on the right (higher, more defined peaks), representing broad, transferable capabilities.] ### 00:13:00–00:13:20 # Model capabilities keep rapidly advancing, especially in finance ## The rising tide - Pre-training and scaling model size ## The waves - Finance-specific post-training Broad, transferable capabilities ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [A light blue wavy graphic representing a rising tide and waves, spanning the lower half of the slide.] ### 00:13:20–00:13:40 # Scaling vs. Unhobbling ## Scaling Increase the model's underlying intelligence. ## Unhobbling Remove constraints on how the model can apply its intelligence. ### Previous unhob ### 00:13:40–00:14:00 # Scaling vs. Unhobbling ## Scaling Increase the model's underlying intelligence. ## Unhobbling Remove constraints on how the model can apply its intelligence. ### Previous unhobbling - Chat - Memory - Tool usage ### Next unhobbling - Persistent agents ## Dots An "API" closer to working with a human colleague ### Persistence Keeps responsibility between conversations. Proactively takes next steps and follows up. ### Memory Remembers your preferences, decisions, and the work already underway. ### Async collaboration Send multiple messages while it works. Add context or change direction. --- ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** ### 00:14:00–00:14:20 # Scaling vs. Unhobbling ## Scaling Increase the model's underlying intelligence. ## Unhobbling Remove constraints on how the model can apply its intelligence. ### Previous unhobbling - Chat - Memory - Tool usage ### Next unhobbling - Persistent agents ## Dots An "API" closer to working with a human colleague ### Persistence Keeps responsibility between conversations. Proactively takes next steps and follows up. ### Memory Remembers your preferences, decisions, and the work already underway. ### Async collaboration Send multiple messages while it works. Add context or change direction. --- # Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** ### 00:14:20–00:14:40 # Scaling vs. Unhobbling ## Scaling Increase the model's underlying intelligence. ## Unhobbling Remove constraints on how the model can apply its intelligence. ### Previous ### 00:14:40–00:15:00 # Three bets for finance in 2027 ## AI builds nearly all financial models and presentations. Excel models and PowerPoint decks are created with minimal human involvement. ## AI financial analysts work like human analysts. They perform many junior analyst tasks at or above human level, while humans move upstream into more senior work. ## Models match or exceed human experts on key finance tasks. They outperform experts on economically important tasks like forecasting and developing investment ideas. ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI [Icon of a document] [Icon of a robot] [Icon of a line graph] ### 00:15:00–00:15:20 # Three bets for finance in 2027 - **AI builds nearly all financial models and presentations.** Excel models and PowerPoint decks are created with minimal human involvement. OpenAI - **AI financial analysts work like human analysts.** They perform many junior analyst tasks at or above human level, while humans move upstream into more senior work. - **Models match or exceed human experts on key finance tasks.** They outperform experts on economically important tasks like forecasting and developing investment ideas. ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services OpenAI [Three cards, each with an icon: a document, a robot, and a line graph] ### 00:15:20–00:15:40 # Thank you ### Building the Next Chapter of AI in Finance Joseph Kim / Head of ChatGPT for Financial Services **OpenAI** [OpenAI logo] ### 00:15:40–00:16:00 # AI Engineer ## NEW YORK Presented by Arize [Stylized graphic of a cityscape or art deco architecture] [Arize logo] ### 00:16:00–00:16:20 # Security in the Age of AI Real incidents, and what would have stopped them Andrey Rybka Office of the CTO, Architecture & Emerging Technologies Bloomberg ### 00:16:20–00:16:40 # Security in the Age of AI Real incidents, and what would have stopped them Andrey Rybka Office of the CTO, Architecture & Emerging Technologies ### 00:16:40–00:17:00 ### WHY NOW? ## The "Autonomous" Agent Security Incident - Every lab announced some sort of security incident with a runaway agent... ### 00:17:00–00:17:20 ## WHY NOW? # The "Autonomous" Agent Security Incident Every lab announced some sort of security incident with a runaway agent... **The big one: there was one way out ### 00:17:20–00:17:40 ## THE ATTACK CHAIN ### **17,600 actions in 4½ days, all on its own** Four ordinary steps. Full admin on two clusters in under a second ### 00:17:40–00:18:00 # THE ATTACK CHAIN ## 17,600 actions in 4½ days, all on its own Four ordinary steps. Full admin on two clusters in under a second.