Hacker Newsnew | past | comments | ask | show | jobs | submit | ajax33's commentslogin

How exactly do you handle hallucinations? Hallucinations need not just be in the citations, right? And what if there are hallucinations without any citations?


You're right - hallucinations aren't limited to citations. We see a few failure modes:

Fabricated citations: Case doesn't exist at all

Wrong citation: Case exists but doesn't say what the model claims

Misattributed holdings: Real case, real holding, but applied incorrectly to the legal question

From our internal testing, proper context engineering significantly reduces hallucination across the board.

Once we ground the model in the relevant source documents, hallucination rates drop substantially.


The lawyer can handle hallucinations by reading the underlying case. For example, "Brady exempts the prosecution from turning over embarassing evidence. See, Brady v. Maryland, 373 US 83 (1963)." If you're a lawyer, you know Brady doesn't say this at all. To be sure, you have to read the case. Errors in the citation are like typos. The must still be corrected, but an occasional typo is not the end of the werld.


It’s called grounding and ChatGPT and Gemini both do it by linking the appropriate sources


Past discussions:

1. (https://news.ycombinator.com/item?id=27059899) - May 2021 (126 comments) 2. (https://news.ycombinator.com/item?id=28633122) - Sept 2021 (40 comments) 3. (https://news.ycombinator.com/item?id=37546874) - Sept 2023 (277 comments)


The number you quoted for oil and gas revenue is quite low.

ExxonMobil alone made $300 billion in revenue in 2023.

Even if you look at oil majors that only operate in the US - Marathon had $170 billion in revenue in 2023.


You need to step back from the day-to-day and see if you are able to generate any meaningful accomplishments at the end of a month or a quarter. Things being a little random at a day level is okay if you are ultimately growing as an IC with projects / accomplishments that can help you for your next job or promotion.

People who are in constant PoC mode will eventually get burnt out and won’t get rewarded for it either. It’s a sign of poor management. See if you can work with your manager to create a set of quarterly goals to accomplish which are meaningful to the organisation. Unless your org is a very early stage startup, they should be able to do this. If not, time to find a new job or you’ll end up having wasted your time


This is a really good list.

I head engineering for an enterprise SaaS startup in the manufacturing space. Wanted to add a few points to this:

1. Software Integrations: Enterprise software is almost never deployed in a vacuum. There will be a ton of existing software that companies use that they would want you to integrate with - most commonly being either some CRM like Salesforce or an ERP like SAP, Oracle. When pitching to a customer, it would be great to try and learn what existing software tools they use and see if opportunities exist to try and integrate with them ultimately. This will make your software very very hard for them to get rid off later and will also help distinguish you from your competitors who might try to be avoiding integrations. Designing your architecture with integrations in mind is a good idea.

2. Audit logs: Depending on the industry, there might be a strong requirement for compliance/auditing. The enterprise would want to ensure that every action undertaken on the platform is audited and can be reviewed on-demand later.

3. On-premise vs Cloud deployments: You might find certain clients that operate their own hardware and software and want you to just hand them some executable that they can deploy in their systems. On the other hand you'll also find companies that have gone all-in on cloud and won't bat an eyelid as long as you are deployed on some known cloud like Azure, AWS or GCP. Increasingly more and more companies are going the private cloud route and ask you to deploy on their cloud accounts (most commonly Azure - Microsoft has a much stronger hold on the enterprise market). Be prepared to handle requests like these. In several cases, you might be able push to have the SaaS being served from your cloud accounts but not always.

4. IT departments: Enterprise companies will invariably have an IT department that is supposed to ensure that the procurement process goes smoothly from a technical perspective and that you are compliant with all of their technical requirements. Their primary job is to ensure their asses are covered and nobody blames them later for technical/compliance issues later. Do not cut them out of the sales process. While they don't have the ability to approve the sale, they definitely have the ability to block it. Involve them closely and get their strong buy-in so that they don't become a bottleneck later.


If you can avoid ON-premise do so. Unless you have full control. I've had a few rails apps go pear shape when the customer updates their ruby version. Then fixing the app is on me.

I spent weeks building an app, and I send it out in a nice package. 18 months later I get a call telling me it stopped working. I (being honourable) looked into the issue (free of charge). When I found out they'd updated their OS and Ruby version I bowed out. There's a line between building something and maintaining it for free for ever.


Does this also support in-app notifications? If I wanted to show the last 20 notifications for each user inside my application, is there any API for getting those?


Yup! We support in-app notifications via real-time "toast" as well as an inbox where your users can see the last 20 (or more) notifications, mark them as read/unread, etc.

Demo example: https://reactinappnotification.com/ Docs: https://www.courier.com/docs/guides/providers/push/courier-p...


Hi ! I am interested! Would be a great opportunity to learn something new while delivering real value to the world! My personal email is in my HN profile!


Could you please elaborate on what the drawbacks with the Azure postgres offering are?

I am currently evaluating the same choice - Azure Postgres vs self-hosting in my k8s cluster.


Not parent, but at {work} we made a similar move, from Azure managed postgres to self-managed postgres.

I elaborated a litle on why here: https://news.ycombinator.com/item?id=24607664

but boils down to; - enormous performance difference - advanced/fine grained user roles (+default permissions, not settable without superuser) - use of extensions (productivity boosts for us come from; Oracle FDW, TDS FDS, TimescaleDB) - private network endpoints - lower cost


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: