This may be a naive question. But is there a way to embed a chat bot like this that only queries the data we feed it, and not the universe of other stuff in gpt? Like I don’t want people using the chatbot in our product to query a good strawberry shortcake recipe. We just want them to query about data we allow it to query which is native to our business.
When you're building applications on top of LLMs, there are a number of central problems that you're trying to solve and this is one of them. Solutions are numerous and widely variable, everything from basic regex parsing to fine-tuning validator models to new programming/modeling languages. Here's some examples:
There’s no foolproof way to do what you want at this point. You could have a separate model trained to infer whether a query is relevant to your product, and then reject the query if it’s predicted to be irrelevant. That’s not 100%, though.
I don't think you can really 100% prevent it. even openai has issues with gpt responding in an undesired way (google Dan, where people try to hack instructions to get responses that are undesired by the openai team). However I think you can make it more difficult (as in the person trying to misuse your chatbot will need to put in some effort to get the strawberry cake, if you have instructed it before to only give information about your product)
Look into https://github.com/NVIDIA/NeMo-Guardrails and specifically to your question there are "topical rails" to ensure the conversation stays on a set of topics you greenlighted.
Also takes care of jailbreaks and allows custom conversation flow templates.
I'm curious how that works, as the documentation is a little under-specified. It seems like it requires specifying exact "utterances" from the user, but I don't think that can be the case -- wouldn't it be flatly useless that way? But it's not clear how to use it to, for example, disallow talking about politics. Or to disallow talking about topics unrelated to the dev's product, for that matter.
This is somewhat possible. I've created a way to chat to our company's material publicly. We used a lot of prompt engineering and custom guardrails to achieve this. However, it severely limited the length of the conversation that a user can have.
I’m shocked at all the suggestions to get a lawyer as step one. Yes you will need a lawyer at some point to get the terms into writing but what you need is business advice to prepare you for a convo w your employer.
That sort of advice is hard to provide without thorough context but few things to think about:
1. How key are you the sales process? If technical due diligence is a big part of the sales process then you have leverage.
2. Are you a key employee that the acquirer will want to put under contract for a couple years to ensure continuity? If so, you have leverage.
3. Do you generally get along w your employer? If so, you have soft leverage.
As an earlier commenter said, it’s all about leverage, leverage, leverage. No (or almost no) owner is going to give you a significant portion of his or her windfall because it’s “the nice thing to do”. This will only get done if you have leverage and the courage to use it.
P.s. there are all sorts of expensive tax implications here too when it comes to ownership transfer. Lawyers and accountants will be needed but not until you’ve neared some agreement in principle.
Much of the marketing automation we do now flows through a single system which tracks user activity end to end with minimal integration work. For example, this means that if a user signs up to receive a white paper it's much easier for us to customize the messaging based on their individual history. It can be done by integrating separate systems together but it gets messy.
I should add that I think we benefited hugely by doing things "the hard way" to start. By learning each and every digital marketing discipline separately we were much better equipped to deploy a big box solution like Hubspot in the end.
I'm more than happy to go deeper on this if you wish. Ping me at jsherry at cbinsights dot com.
Surprised this article mentioned chat only once. The secret sauce to Bloomberg's stickiness is their collaboration features via what's know as Instant Bloomberg.
I've got a Symphony account. It's okay I guess, but Bloomberg has the network effect, and that's a very difficult thing to overcome because for at least some period of time you'll be asking people to use multiple communication platforms, at which point, some of them will just revert to their original platform.
I would have to guess once Symphony is proven some of the banks will strongly encourage/require communication with their traders to go through Symphony. or it might be another failed Bloomberg-killer. Will be interesting.
I don't think the first commercial Bloomberg terminal was in 1982, but I could be wrong. Michael Bloomberg was still at Salomon until mid-1981. Maybe there was a prototype sometime in 1982. Merrill didn't invest until 1983 and that's where the early terminal got all the bond data. I don't think it was a big factor in the market until like 88-ish, Salomon launched their Yield Book project to compete around 1988-89. There was some kind of an exclusive with Merrill and/or bad blood with Salomon and they couldn't even buy a terminal.
CB Insights | Lead QA Engineer | NYC | Onsite | Visas Welcome
CB Insights is seeking an experienced QA engineer to build out our automated testing suite and ultimately build a team to own all aspects of QA.
By way of background, CB Insights is a National Science Foundation backed software-as-a-service company that uses data science, machine learning and predictive analytics to help our customers predict what's next—their next investment, the next market they should attack, the next move of their competitor, their next customer, or the next company they should acquire.
The world's leading global corporations including the likes of Cisco, Salesforce, Castrol and Gartner as well as top tier VCs including NEA, Upfront Ventures, RRE, and FirstMark Capital rely on CB Insights to make decisions based on data, not decibels.
If interesting, please contact me (Jonathan) directly at jsherry@cbinsights.com.
If you have a taste for stories that span the spectrum of airport carpeting fanboys to those with more historical significance such as the story here, I cannot recommend the 99% invisible podcast enough.
To list a couple others off the top of my head: SF's public stairways, NYC's cow tunnels, sound design for televised sports, Plimsoll lines, elevator brakes, architecture for deaf people.
I don't know if I'd call it a "spectrum" so much as a "crazy pile of everything design related." Great show.
In NYC, GA conducts nighttime Meet & Greets with their "graduating" classes. As an employer, this is an efficient way to recruit. I attended one a couple weeks ago and thought it was fairly well organized. If GA can help people get placed into companies, then that's a huge advantage / differentiator for them versus online-only course providers.
So without speaking the quality of the classes themselves - GA specifically, or the concept of online v. in-person - this is a perspective worth considering before slamming the in-person education model altogether. Execution of the actually classes is obviously a separate discussion, but as I said one I'm not prepared to address.
The one here in DC is already doing an excellent job of matching people up with start-ups and non-profits. We've also got NGO's like the World Bank showing up.
My view is this: if a class has good teachers, a difficult, portfolio driven curriculum, and a decently high-bar of entry (my DBA coworker struggled to be admitted to the course because he was rusty in his programming skills. He ended up having to put in a lot of time just to be able to pass their test to gain admission into the course), then it will probably be a good course to take.
Veeva Systems is another that flew under the radar. B2B Life Sciences software company that raised $4M five years ago from Emergence Capital - a stake which was worth $1.2B at the time of their IPO in October.(1)
Good on Dan Primack to shed some light on the "boring" side of VC.
Jonathan from CB Insights here. We published this report in partnership with Silicon Valley Bank and Orrick. For startups, investors and advisors interested in the NY ecosystem, we hope it is helpful.
do you guys track companies in the energy sector specifically? It's hard keeping track of M&A in that sector and would love to have some data that consolidates that kind of info and other metrics like you have in the almanac report.
Terrible headline for sure. But in sushi restaurants, this mislabeling is actually very common. More often than not, escolar is labeled white tuna. You know it is escolar when it's a pure white and opaque fish with a buttery taste, which although mislabeled, is not at all unpleasant. Rumor has it that of you eat too much, it also gives you diarrhea but fortunately I've never had enough of it to test that theory.
As for real white tuna, there is such a thing and it's called albacore. Aside from being found prevalently in cans, it can also be had at many finer sushi restaurants and it's delicious. White but lightly translucent and often served with a bit of horseradish.
Is this feasible? Thx.