Jekyll2025-12-18T18:18:07+00:00https://josephharrisonlim.com/feed.xmlJoseph Harrison-LimJoseph Harrison-Lim's personal website.Joseph Harrison-LimMobile Browsers: Overflow vs Sticky2024-10-22T04:00:00+00:002024-10-22T04:00:00+00:00https://josephharrisonlim.com/blog/2024/10/22/mobile-browsers-overflow-vs-stickyA common thing you want on sites is to have a header that is always at the top, so that users don’t have to scroll all the way up the page to reach the header’s navigation.

One thing you can try is setting the body to overflow: hidden, and then putting overflow: auto on the content under the header. This will make the header stay in its spot, while the content underneath can be scrolled.

This works well on desktop browsers (including desktop browsers that are in “mobile mode”), but on mobile browsers, it doesn’t work as expected. Instead, you can’t scroll fully. This is because the mobile browser includes its own header and footer in height calculations, so you can’t actually scroll to see the bottom of the page.

Overflow

If you instead use position: sticky on the header, it will stay in its spot, and you can scroll to the bottom of the page.

Sticky
]]>Joseph Harrison-LimOpenAI API Rate Limit on First Request2024-10-21T04:00:00+00:002024-10-21T04:00:00+00:00https://josephharrisonlim.com/blog/2024/10/21/openai-api-rate-limit-on-first-requestI decided to try out OpenAI’s API for the first time yesterday. From what I saw within the developer dashboard, it seemed like I would be use it without paying anything up until a certain token or request limit. It seems like this was not the case. If you make an API request with the token you generate and you haven’t added any payment information and credits, you will always get a 429 error.

Ultimately it’s cheap enough to start with so I don’t mind, but worth sharing in case anyone else runs into this.

If you want to see what I’m using it for, check out total-thought. In Total Thought, users are able to write out topics to discuss, such as “Should I use the gym or build my own home gym?”, write the pros and cons of each (others can also contribute!), and then crowdsource upvotes on each pro and con. I am using gpt-4o-mini to summarize all of this data and try to give a reasonable suggestion on which one to choose.

]]>Joseph Harrison-LimGetting from Deep Learning to LLMs2024-10-18T15:58:00+00:002024-10-18T15:58:00+00:00https://josephharrisonlim.com/blog/2024/10/18/getting-from-deep-learning-to-llms5 years ago, I felt fairly in tune with the ML world. I understood how to solve classification and clustering problems. I learned about neural networks and the process of deep learning, and I tried writing my own feed-forward network to showcase it. Of course this is a little bit ironic, because 2017 is when the Attention Is All You Need paper was authored.

Fast track until now, and I’m completely behind. LLMs are a big change that happened and I didn’t know what led to its fruition. I wasn’t sure how neural networks fit into the picture anymore. Luckily I was recently recommended a 3b1b video which explained this to me very well. This channel released many useful videos which I watched those 5 years ago, and they taught me how key concepts worked such as back-propagation. Early this year they released follow up videos to connect the dots.

What I learned is that the feed-forward neural network is still relevant; it is now the multilayer perceptron inside the transformer. In fact, there are multiple multilayer perceptrons inside the transformer. The big new piece of the puzzle is the attention block mechanism. Again, the 3b1b video explains it very well, much better than I can do via markdown. To summarize anyway, it does some mathematical tricks with dot products to find relevance from one word to another, and this is represented via the dimensionality of the vector produced.

This bit of catch-up was a great relief. I feel primed to learn more, and hopefully start developing with this stuff soon.

]]>Joseph Harrison-LimJekyll, to Gatsby, to Jekyll2024-10-17T00:25:03+00:002024-10-17T00:25:03+00:00https://josephharrisonlim.com/blog/2024/10/17/jekyll-to-gatsby-to-jekyllOriginally I had used Jekyll to create my personal website. I wasn’t acquainted with Ruby at the time, and still, it was simple to learn and manage. Eventually I discovered Gatsby. Gatsby fills a similar purpose. Both are static site generators that make it easy to host a site through GitHub pages. Gatsby is based on Node.js rather than Ruby, and lets you use React and GraphQL. This sounded really good to me at the time since I was familiar with the stack, and it was easy for me to add some code for page transitions and animations, especially with react-spring.

Fast-forward 4 years and my priorities have shifted. I wanted to update my site again, however the maintenance cost was too high. I had an overwhelmingly large amount of packages that needed major version updates, and it was overbearing to look at. This isn’t the type of project I wanted to have for my personal site, so I switched back to Jekyll.

With Jekyll, I don’t have to worry about this. Jekyll doesn’t require many gems to run, whereas Gatsby required multiple packages from all of the plugins. Gatsby would of course have me install the world through node_modules, whereas Jekyll requires very little. The Gemfile.lock is under 200 lines! For a casual project with a low update frequency, this is far more bearable. I will happily sacrifice my beloved TypeScript, React, GraphQL, Et Al for simplicity.

To my surprise, GitHub Pages can also now automatically deploy jekyll sites without us needing to write any CI YAML files ourselves!

]]>Joseph Harrison-Lim