I’ve worked across a variety of roles and across several top banks and have been involved in the use of AI in banking for over a decade. I’ll be here from December 4 to December 11, answering your questions intermittently.
5 Likes
Hello @Silverleaf , thank you for taking the time to do this.
I just wanted to ask what you think of the current “mainstream” AI models (mostly Gemini 3.0 & GPT 5) and whether they have much use (as they are) for banks.
Separately, which is your favourite LLM? Why?
I’ll assume you are referring specifically to Gen AI models… The technology divisions of banks are focusing their AI efforts to reduce operating costs, namely with Gen AI and Agentic AI. They have a hard time distinguishing between automation, RPA, & AI, and have not yet thought of the comparative cost of AI agents in run-time environment. That being said, current “mainstream” AI models, once approved for internal use, are typically used for question answering, and at best used for agentic RAG. They are typically poorly adopted and under utilized due to performance issues (latency, reliability, etc.). While these large language models may not provide the cost/benefit expected… smaller language models, often fine-tuned and self-hosted, will, offer better cost-benefit as they would be better tailored to meeting our domain specific use cases… especially where regulatory requirements are critical.
My preferred LLM is Claude, due to its lower hallucination rates and Anthropic’s commitment to responsible and safe AI development.
5 Likes
What’s the appetite of using AI at compliance department? I mean do they have hard concerns or just they wait to see how it’ll go etc? Do they worry about data safety etc?
1 Like
To what extent do you think the use of LLMs is being pushed by management who don’t really understand the tools, and who just want to automate everything and cut costs? LLMs have a place, but I’m not sure they’re as useful as the C-suite seems to think
Has there been much consideration in finance of the environmental impact of AI, particularly generative AI? Are the more traditional AI use cases in banking similarly bad for the environment, or is that only a GenAI problem?
2 Likes
Hello @Silverleaf , your post and this forum is coincidentally exactly what I needed. After 15 years working in investment banking and then wealth management in Switzerland I have enrolled in a training program in Python Data science and AI applied development. My goal is to be able to do what you are doing. Do you think its possible ? What is your background and what specific knowledge do you recommend I acquire ?
1 Like
Very high, actually. Compliance typically involves heavily manual processes and reviews of extensive documentation. This often requires high human capital costs. While Gen AI lends itself well to these types of processes, there is little room for error in misclassification… so you will see supervised models used in conjunction to provide a certain level of assurance in accuracy. What there isn’t as much appetite for is the risk that would be introduced by full autonomization of compliance processes with agentic AI; however, some places are introducing agentic compliance pilots incorporating HITL (human in the loop).
2 Likes
There was certainly awareness of the rising cost of compute associated with AI workloads, which is tied directly to data center carbon emissions and impact to local water tables and resources in the communities where data centers are located. Prior to Trump’s 2nd election, efforts had begun to begin tracking carbon emissions for these workloads and techniques such as quantization, distillation, model compression were being explored for both the finops benefits as well alignment with carbon reduction commitments. With Carbon commitments now being rolled back, it’s gone on the backburner. However, rising costs and improvements in performance from much cheaper models using these techniques has piqued interest again.
As far as traditional AI… any compute-intensive workload will have environmental impact.
1 Like
I’ve been in AI for close to 15 years and have a physics background. Years ago I would have recommended studying advanced stats, partial diff eq.s, software engineering, etc. While it is still good to start with the fundamentals, I don’t believe it’s a hard requirement anymore, and there are many different types of roles in AI now. If you want to be an AI researcher working on foundational models… yes start with fundamentals. If you want to be a data scientist or ai engineer who consumes these models and then does fine tuning, etc. , a data science and software eng. education is sufficient. You can also be an AI product manager, a data engineer, work in ai business development, ai governance, and so on…. which do not require development skills. I’d explore the different new roles in AI and decide which path makes most sense for your interests and skills.
1 Like
As a student just embarking on the financial services journey and full of passion for markets and trading, what do you think I should do to ensure my skills (finance degree, CFA level 1) get me a job in the AI enabled future. Thank you for your help
1 Like
Culturally, how do you think the lower/mid-level people have changed their attitude towards AI staff since the rise of GenAI? Have you faced any more or less friction from them as more effective AI tools have been introduced, or is the difference negligible? I imagine there’s probably a mix of some people happy that your tools have made their life easier and some people upset that your tools may make them redundant.
1 Like
Thank you for answering our questions here. I would be interested in becoming an AI product manager. Can you explain what this job involves and how I would need to train for this?
1 Like
How do banks typically go about trying to encourage adoption? Do they offer role based training or just say “use it”?
Morning, thank you for your availability.
I would like to ask you, concerning the AI potential. How, about you, the right equilibrium between “AI-Driven”, and “AI-Assisted”? Leaving by a definition of the “good innovation”, as the “innovation capable of combining new services with market integrity, security, and investor protection”, is possibile to define a border for this discipline in every Continent?
Regarding the major risk of fraud in “transaction banking”, where the major risk, in Europe or in United States? I ask in legal and reputational risks, and about statical terms.
I remain at your disposal.
Giovanni D’Avanzo
Something as simple as setting aside a weekend to try building an mcp server can you give a level of insight most don’t have into where and how best to apply these emerging AI tools to your job.
You should try out different available models and see how they perform against different tasks that are of interest to you, this will give you better intuition on which models work best for different tasks you need AI assistance with.
Building up and maintaining your modern AI tool literacy will be best achieved with hands-on practice and experience… and it will put you miles ahead of the majority of people who will have no idea how these tools might be beneficial for their daily tasks and jobs.
For some, it has made the relevance of AI to their daily lives real. However, the fear of redundancy began many years before the rise of GenAI. The degree to which people are happy about the new tools are directly related to the efficacy of their operationalization within the enterprise. For something like github copilot, which has been rolled out in advance of powerBI copilot….. you see the difference in app developers’ reaction to AI (enthused) vs. report developers (less enthused). It’s just a matter of ‘how practical and useful is this for me right now?’. Some people argue (and are frustrated) that redundancies have been executed far in advance of operationalization of relevant AI tooling…. I’d agree.
You’re typically translating qualitative customer and consumer needs into prioritized features and functionality for an application, and tracking progress against their expectations with quantitative measures. Then, working with an engineering manager who then translates this into technical requirements and a realistic timeline and effort/cost estimation.
Probably working in management consulting in the AI sector would be a good foundation. I’ve also had engineers who went and got their MBA get into this field.
Hello,
I am a student in the UK.
I often see advertised part-time courses from universities, usually around 8 weeks long, promising an entry-level course in AI for business, or AI in general, for relatively reasonable prices. Some of the universities that offer these courses are reputable - Oxford, for example.
My question for you (and thank you for taking the time to do this AMA) is this: are any of them worth a damn? If so, which ones? Or are these all just money grabs?
Thank you!
1 Like