I'm happy to say AntConc 4.3.0 for Win, MacOS, and Linux is now released, including the new ChatAI feature from recent beta versions. ChatAI allows you to interact with various LLMs and get insights on traditional corpus results. I hope you like it! https://t.co/dJSr51yvxv

Can't believe one week has already passed since the end of the @CL2025Conf. This year, I had the great honor of serving as one of the plenary speakers, giving my talk on AI and corpus linguistics in a really lovely venue. Big thanks to all the organisers. #CL2025

Our paper on AI applications in vocabulary testing is finally out!

Arriving early for the start of the Microsoft AI Tour here in Tokyo. A great way to finish my sabbatical year, where I focused my research, talks, and app development on this revolutionary technology.

A bit late, but I’m currently in Erlangen, Germany, enjoying the first day of a @rc21project.bsky.social symposium. It was a great honor to open the event with a talk on AI integration with concordance methods, focusing on new #AntConc functions.

Happy to announce our paper on vocab testing using AI was just accepted by the CALL journal. My first collaboration with @Jeffrey51122028, @AaronOlafBatty, Keita Nakamura, Christopher Nicklin, @StuartM13041063, and Kanako Tomaru. The preprint is here:

Today was my first day back at @waseda_univ, where I had the pleasure of hosting Nathan Wang, the CTO of the educational AI company @EdexiaAi (YC W25). Very much enjoyed discussing the role of AI in education and hearing about the company's plans.

Back in Japan after a fantastic half year sabbatical at UQ. Good news is that I'll be continuing my connection as a visiting scholar until 2028, serving as a consultant and developer for the Language Data Commons of Australia (LDaCA) project. https://t.co/CCHnRJhGqH

Last day of my sabbatical at the University of Queensland. I’ve had such a great time in Brisbane. It’s a really lovely city. Huge thanks to @ronautic and @DrCrosthwaite for being such great hosts, and to everyone else I’ve met during my time here for being so welcoming.

Very much enjoyed attending the Brisbane AWS User Group this evening. Great talk by @markpromnitz and a great atmosphere in the AWS offices. Nice getting to know Bill McKeague, too.

For anyone interested in my recent seminar on running DeepSeek-R1 and other open-source/open weight AI models locally, I have now created a public GitHub repo that includes all the slides, demo files, and a complete tutorial. I hope you find it useful! https://t.co/1IeLRLbWve

Thanks to everyone who attended my presentation today at the Queensland Govt. AI Community of Practice. The topic of running DeepSeek-R1 and other AI models seems very timely as we had a huge turnout—300 people! I'll be making the slides and materials publicly available soon.

For anyone with a .gov.au email, I’ll be giving a talk on “Local, private, portable AI: Getting started with DeepSeek-R1 and other open weight models” on Feb 10 (Mon), 2025 for the QLD Govt. Customer and Digital Group (QGCDG) as part of its AI Community of Practice series.

Happy to announce that I’ve just started a new contract position as an Educational AI Advisor for the Queensland Government. Many thanks to Marco Fahmi (@dataronin) for setting this up. For anyone wondering, I'm still also a full-time professor at Waseda University in Japan!

Hi everyone. Here's an event that I'll be presenting at in March 2025. I'll be talking about "AI-Assisted Concordancing: From Search Support to Pattern Profiling" where I'll demonstrate AI models runnning locally in #AntConc (no OpenAI account needed). https://t.co/2UoaI3bRuZ

Hi everyone. It's a big honor to be giving a plenary at the CL2025 conference this year. If you're thinking of going, the deadline for regular submissions is Friday, Jan 17, 2025. Maybe I'll see you in Birmingham!

I have to give a huge thanks to @karpathy. His video series explaining micrograd really is the best out there if you want to understand the calculus behind these models.

Of course, this is just a toy example, where I trained the LLM on a mere 18 tokens (3 sentences)! But, there's actually a lot going on here. For example, the model can scale like ChatGPT to trillions of parameters and learn from the entire web. I would just need a lot more GPUs!

One goal of my sabbatical here at UQ in Brisbane was to build an LLM from scratch so I could be sure I fully understood every component and could experiment on each. I think I'm final there. Here's the first toy example.... notice how the model has learned that cats sit on mats!

Hi everyone! I've just completely updated my website to make it run quicker and hopefully easier to navigate. I hope you like the changes. If you notice any missing or broken links, do let me know. https://t.co/61C6HFvrWA

I should note that I created custom source material for these, and I also listened back to the recordings to make sure everything made sense. I know that some people worry about AI content, so I'd love to hear your thoughts!