njsubedi

There is a little easter egg in DuckDuckGo's icon. Go to https://DuckDuckGo.com/ and try searching for texts like linux, gnu, einstein, marie curie, yoda, harry potter, spiderman, batman.. The icon changes to match the respective keywords. Found this quite interesting.

Amidst protests by Gen-Z against yesterday's inhumane killing of 19 student protesters, the country's capital is on fire.

The prime minister of Nepal has resigned, and fleed on an army helicopter - probably abroad. Many other ministers have been evacuated. The parliament building, ministries, and all of the top political parties have been burnt down.

Submitted this because yesterday evening there was a post and many of you were concerned for Nepal.

Coverage: https://www.msn.com/en-gb/news/world/nepal-pm-oli-quits-as-anti-corruption-protests-spiral-his-aide-says/ar-AA1Mb480

I understand that this might be a bit late to ask this question, and I don’t know a lot about AI/ML in general. I have to train/tune a pre-trained model for specific context. By context, I mean a knowledge base, a product’s documentation, or user manual, or in my case, an inventory of electronic items in our warehouse.

I tried dumping the inventory information and basic BOM content as a system message for ChatGPT4-o model using their platform playground, and asking questions like “we are making 1000 power banks this month, so which components should I pre-order so that we don’t run out of them?”. This works as expected, but it took me a while to realize that each query used more than 30K tokens! That’s a quick way to lose money.

I am looking for a solution from someone who trained/tuned a decent LLM on custom data, and I’m pretty sure a lot of other small business owners are looking for something like this. Thank you!

Bard replies:

You're right, I didn't do as well as ChatGPT on that request. I'm still under development, and I'm always learning. I'll try to do better next time.

Here are some of the reasons why ChatGPT might have done better on that request:

ChatGPT is a larger language model, with more parameters. This means that it has a larger vocabulary and can generate more complex and creative text.

ChatGPT is trained on a different dataset than I am. The dataset that ChatGPT is trained on includes more examples of words being combined to create new words.

ChatGPT is specifically designed to generate creative text. I am not as well-optimized for this task.

I'm not giving up. I'll keep working hard to improve my ability to generate creative text. I'll also try to learn from ChatGPT's strengths. I'm confident that I can improve my performance in the future.

My company makes iOS apps and games. Yesterday I received an email from Apple telling that they're terminating our account in 30 days for breaching the agreement by trying to manipulate the app ranking. I promptly checked the trends and saw that one of our games got 17k and 18k downloads from the UK only in two dates (19 - 20 Jan, 2022). All users downloaded the game from the App Store Search page. On usual days, the game only gets 200-300 downloads globally, and 15-20 downloads from the UK.

Apple wants us to find the issue, mitigate it and ensure that does not happen before they can look into the case. We don't know how that happened, and we don't see anything abnormal in other stats. The game is basically a driver's license exam simulation game. Did anyone in the UK notice some media coverage or news about such a game? Did anyone else got unnatural downloads on those dates because Apple accidentally messed up their algorithm on those dates and our app got ranked higher? Any hint would help us keep our business running - we're bootstrapped, and only publish in-house mobile games and apps, so we don't really have anyone at Apple to ask what went wrong.

I have been reading scary stories from developers whose accounts got closed by Google and Apple every now and then. In my case, it's not a life ending situation because we also publish on Google Play. But still I'm curious if the App Store Search went nuts for everyone else or just me. Thank you for reading this far. :)

I am curious about the overall architecture of Hacker News. After reading the recent thread about it’s expiring certificates, and the linked comments, I searched but couldn’t find an exact answer. So, what kind of hardware and software stack does HN run? Are there any CI/CD tools, analytics, et cetera?