👨🏾‍💻 Ben Ilegbodu

When did senior engineers start acting like every task needs to be fully scoped for them? Lately, I’ve heard devs say they want “clean Jira tickets.” Let’s be honest about what that really means: No ambiguity. No product thinking. No ownership. Just code. Preferably handed to them in a neat little box. If that’s the kind of work you’re looking for, there are places for it: internal IT at a bank, government contracts, legacy systems in maintenance mode. But if you’re working in a product company—especially a startup or scale-up—that posture won’t get you far. The job is messy. The requirements do change. The customer will contradict themselves. That’s why we need engineers who lean into the ambiguity, not away from it. This is exactly why I wrote Product Driven. Too many teams are optimizing for clean tickets instead of customer outcomes. But product thinking isn’t a “nice to have.” It’s the job. I break down what that looks like and how to lead for it in the book. 👉 https://buff.ly/1W4AoCz

1,751

445 Comments

Creating SDK documentation that truly enhances the developer experience requires following a set of established best practices. These principles extend beyond just writing; they also involve the design of the SDK, the quality of the examples provided, and the overall documentation process. My latest PDF outlines these principles for creating effective SDK documentation. Take a look to learn more about the best practices for SDK documentation.

31

Post-train once, reuse many times! Up next in my "Why Continuous Pretraining is Awesome and Its Time May Have Come" series: insights from the area of model merging. I've posted about model merging a few times before, so I'll link those in the comments below if you need to get up to speed. With continuous pretraining, you take a base model (one without any post-training: instruction following, safety training, etc.) and pretrain it more on a new dataset. This could be to update its data cutoff period or to transfer it to a new specific domain. Then you have to reapply your whole post-training stack on top, or do you? Model merging has uncovered a very useful property of models: that of "Task Arithmetic". We can fine-tune a model and subtract the base model weights, leaving a task vector [1]. These vectors can be combined with models from the same base to add the corresponding capability. For example, if we have pretrained a base model and then instruction-tuned it, we can compute the instruction tuning vector for this model. Later, we can continually pretrain the base on a new dataset and then simply add the instruction tuning vector on top to get the best of both worlds: a domain-transferred model with deep knowledge in a new field and top-quality instruction following. This was demonstrated clearly in [2, 3], showing this transfer of instruction following on both Llama and Qwen models — with [3] achieving 46.9% improvement on instruction-following benchmarks without any additional training. Best of all, this transfer doesn't only apply to instruction tuning but all sorts of capabilities you may want to fine-tune for, as shown with DPO, RLVR, and RLHF on the OLMo models in [3]. Practical implication: Organizations can maintain a single high-quality post-training pipeline and apply it to multiple continuously pretrained versions of their base model, dramatically reducing the cost of keeping models up-to-date. References: [1] Editing Models With Task Arithmetic https://lnkd.in/ekM7VdK3 [2] Balancing Continuous Pre-Training and Instruction Fine-Tuning: Optimizing Instruction-Following in LLMs https://lnkd.in/eUM2PkgM [3] Efficient Model Development through Fine-tuning Transfer https://lnkd.in/e6AuRbQs Shoutout to Alessandro Giagnorio for highlighting some of this excellent research to me!

829

27 Comments

{quarto-webr} v0.4.3 "Bumpity Bump" is a new maintenance release of the Quarto extension that enables interactive R code execution in web browsers. This update bumps to R 4.5.1 by default (via webR 0.5.2). 🔗 Repo: https://lnkd.in/gjSFKBUf #RStats #Quarto #WebR #DataScience

30

I don't write dedicated "unit" tests for every public method or every function. Some might interpret that as not needing high test coverage. Just to be clear, my test coverage is usually 100% or as near as dammit. There's no code that isn't tested.

31

2 Comments

Screw vibecoding. You know what LLM coding agents are *really* useful for? Reverse-engineering legacy code! "Hey can you research and explain how the flux capacitor settings are loaded, specifically where are the margadon-level overrides applied exactly?" It's a lifesaver.

389

101 Comments