Would you, should you, have you deployed GPT4 inside your business yet?

Are people using GPT4 successfully in real commercial products yet? Is OpenAI’s latest/greatest API a good hire?  A recent timely Ask HN discussion hacker news this week poses this question. Go read the whole thing, or here are my meta-take-ways for anyone trying to build with GPT4 right now.

The model can do useful work today
Some interesting feed across a mix of viable usecases: generating marketing copy and sales emails, to “Correcting or filling missing information in structured data” or Correcting or filling missing information in structured data, or data extraction like from websites or documents, or for internally searching for company information.

But… Availability and performance is a challenge
OpenAI’s APIs aren’t always reachable and response times are variable. Do work around managing for retries, or fallback to other processes if OpenAI is not available. Rate limits are a challenge too, and the process for appealing those is going to be difficult with the current level of demand.

Mix GPT4 and 3.5 versions for speed and Cost considerations
GPT4 is the most accurate but also the slowest and more expensive per call. But you can also try mix and match for usecases where 3.5 is good enough or as a first pass. Test and optimize. At MainStreet, we had good success using basic GPT3.5 to generate super-specific customer help and training material. E.g. “draft a help center article on how [general finance concept] might apply to expenses for [specific job role category] in [customers specific business vertical]. We’d generate this kind of content offline, then review it before publishing. Even with manual review, the speedup vs generating a broad set of help content from scratch was enormous.

Keeping human review in the loop or being clear/transparent with your users

Either it seems people are mostly using GPT internally, where quality of output just has to better or more scaleable than a previous process. Or folks are building apps that explicitly expose the AI to their customers, but adding value through a novel UI or domain-specific assistance with prompting. The ‘copilot’ modality when coding or creating is already a proven commercial model and incredibly popular. Will some equivalent work for banking, investing or financial management apps? It will be interesting to see how easily all of this extends into more regulated or professional/fiduciary responsibility domains. But with the right controls, transparency and model-refinement, it will get sorted out.

Key Takeaway: The big cloud vendors will probably make all this better, just be prepared to pay. The business case here is pretty clear for Azure, Google and AWS et all. Offering, enterprise grade availability, as well as data privacy for custom-trained/refined LLM models is going to be huge business. I could also see opportunities major vertical-oriented players offering something similar. Bloomberg has announced their GPT model for Finance. I’d like to see what Stripe or Visa do for models trained on payments, or models for retail banking, lending, accounting, insurance etc.

Relevant links:

Ask HN: Who has deployed commercial features using GPT4?
Open AI API docs
Microsoft Azure OpenAI services
Google Cloud AI (Note Google Bard API is still closed and invite-only)
Amazon AWS Generative AI Announcement (Announced April 13/2022)
Bloomberg GPT (Announced March 30)

image credit: Midjourney “A robot works in an office, ai, paperwork, midcentury modern”

 

 

 

 

Posted in Uncategorized | Leave a comment

Latest ChatGPT Hacks for PMs

As a software product manager it’s hard to think of everything. And you know, in our heart, the true route cause of a lot of bugs is that… the requirements could have been better. This is where I’m kindof excited about he potential of using generative AI also as a co-pilot for PMs.

A couple of prompts I’ve run across that could be of value to a product team in daily work. Of course PMs using these kinds of prompts strictly as first drafts. Or better yet after writing a spec or ticket asking GPT to write a similar ticket and then compare. The tool may give additional ideas or areas/gaps of requirements that are worth adding.

Ticket Writing Prompts:

"You will act as a consultant for tech product managers. Your primary function is to generate a user story and acceptance criteria, given a high level feature description. The user story should be catered to the specific mode of user interaction (e.g. web, mobile), using best-practice UX design guidelines. If you need additional details to provide a good answer, you will continue asking for more context until you have enough to make your user story. Ready to start?"
"As a product manager, I'd like ChatGPT to create Jira tickets for me in the context of a project focused on <enter software description>. For each ticket, I will provide specific information about the bug or feature, and ChatGPT should include this context, along with any other relevant acceptance criteria. As more tickets are requested within the same chat, ChatGPT should remember the context of previous tickets to develop a stronger understanding of the platform over time. I can also provide a list of previous tickets to establish an initial knowledge base. 

First, you should ask me to provide you some examples of previous tickets so you can understand the structure of our tickets and some base knowledge about what has been built already."

Disclaimers!
Remember that anything you put into a chatgpt prompt essentially becomes public and may violate company principles for leaking proprietry information. Instead you can use the API mode, a dedicated azure instance or a 3rdparty tool based on same that will be safer. OpenAI only keeps rights to (re)use your inputs on the consumer public-facing version of ChatGPT.

 

Posted in Product Management | Leave a comment

Top Fintech trends for 2023

I’ve been spending a lot o time recently chatting with other fintech founders as well as my consulting clients on what the near future holds for fintech. Here are a few of the topical trends that keep coming up.

# 1. Of Course Generative AI.
This one’s obvious because it’s clearly impacting the way we work and get stuff done in every industry. There’s so much opportunity in accelerating how we generate customer and internal communications, documentation and code. As a learning tool and as an oracle of (mostly true) knowledge on any subject. But also, there’s real potential to solve longstanding challenges in open banking and open finance. Huge problems with data quality and consistency has been one of the dirty secret problems in fintech for a while. Now with new standards like ISO20022 emerging with (hypothetically) rich new rails for moving payments data, that just creates more problems of how to consistently supply these rails with useful data from fragmented sources. And then on the other ends of the pipes, the parsing and finding any valuable sense-making from firehoses of rich-but-inconsistent payment data, bank data, accounting data etc. You may have seen @Shpigford playing around with GPT4 to fix card txn data. Strong new LLM AIs may also be the answer to…

#2. Plateau of productivity of Open Banking and Open Finance
API aggregators of fintech data (think Plaid, Merge, Codat, Finch, Pinwheel etc.) are hardly new-news. But if you’ve been building on them, the experience is getting steadily better. Quality, reliability and coverage issues are improving. New AI tools might give us another step function in utility of creating new value from all of these sources. But the other big news is that it’s not just ‘read’ usecases anymore. Write-back cases are emerging, and 2023 is the start (read the US catching up to rest of world) of ‘execute’ permissions coming to open banking with new realtime payment rails starting to come online.

#3. RTP and ISO 2022
Realtime funds transfers are happening, finally happening and in the US too. That said, implementations like fednow still have a lot holes and missed opportunities. But at least it’s happening. The new ISO standard is super interesting too for it’s potential to convey SO MUCH more useful context and SKU info with payments. All that potential though, is still going to be just potential for a while now. I think it’s going to be a long while before ISO2022 and RTP is being used and implemented effectively and ubiquitously. But that’s years of opportunity and solution building for us builders to work on, just on the platform/network side. Not to mention the near infinite number of solutions that could/should be built on top of fast, data-rich, global and ubuiquitous money movement rails. Once those conditions are finally met. Final note: Blockchains never were, and never will be needed to make realtime payments work. Computers have been more than capable of realtime messaging since arpanet was invented. Fast payments is a problem of firstly multi-party coordination, standardization, competing economic incentives, politics, risk-controls and regulatory compliance. Actual technolonogy choices as long been very low on the list of the harder problems to solve.

#4. True digital issuance
Have you noticed, that you really don’t need to bring your wallet out with you to pay for (almost) anything? I’ve been waiting years for this tipping point. It means something special. Digital-only issuance (provisioning a card straight to applepay, chrome etc.) becomes… enough. It also means that the marginal cost of spinning up a unique payment credential can approach zero. So may interesting usecases here for spinning up virtual ‘cards’ on-demand in the consumer space, not just in commercial (where virtual cards have been taking off for a while now).

#5. The year of Post-crypto awakening
FINALLY the crypto bubble is bursting. I really lookforward to the clearing of the crypto fog. Once free of that miasma of bullshit, so much talent, time and attention can potentially get back to building real solutions based on actual technology. Still, hold on to your seats for the last few shoes to fall, we’ve yet to see the collapse of some remaining mostly-fraudulent bubbles like the ‘stable’coin Tether. But it’s a matter of time.

#6. The Post SVB cold shower
2022 was a brutal year of valuation implosion across fintech. We might hope that q4 last year was the bottom and that things will trend better. But of course the investment environment will still be challenging and fintech may never again be as (inexplicably?) sexy as it was through the last few years. I am also concerned that we’ll all take the wrong lessons from the collapse of SVB. I’ve seen this movie before in past cycles. A tendency for the big institution and from regulators to retrench from all innovation and creativity in financial services. The tragedy of SVB was that they were a great bank! One of the few that would even provide high quality, essential basic banking services to growth companies and SMB. 90% of the bank was being run exceptionally well. What failed them, were egregious but very specific failings in the treasury department.

Posted in Uncategorized | Tagged , | Leave a comment