Excited to launch DataMap – a portable, browser-based tool for generating heatmaps and PCA & tSNE plots. No installs. No servers. Your data never leaves your device. ✔️ Interactive heatmaps, PCA, and t-SNE ✔️ Transform & normalize data on the fly ✔️ Upload CSV, TSV, Excel + annotations ✔️ Generate R code for reproducibility Ideal for visualizing high-dimensional data matrices, such as RNA-Seq data. It's a Shiny app deployed via shinylive and WebAssembly. Also available as an R package. Try it now: github.com/gexijin/datamap Source code: github.com/gexijin/datamap Preprint: arxiv.org/abs/2504.08875

- The greatest waste isn’t lack of ability, but waiting for a future self. - The brain favors comfort now over rewards later. - Tiny actions break the cycle and build self-trust. I learned these from a video titled "How 90% of People Waste Their Lives", about McGonigal’s book The Willpower Instinct. .... while doomscrolling in bed.

A small trick: "Briefly summarize this as an ASCII flowchart." in ChatGPT, Claude Code, or Codex. It's a quick, visual way to use AI to help understand a code/paper, etc.

I just received 4 free Claude subscriptions for my students through Claude Science. I paid $15 for a premium seat with 5x usage for myself. You need to be a university professor. Maybe they will be harvesting our data for their own science-related projects? You can't really trust these guys. anthropic.com/news/expanding…

Is this the future? Two Cybercabs behind a Waymo yesterday in Austin, TX. Screenshot from TSLA99T's YouTube video.

Should you run an open LLM locally to save cost? I tested Qwen 3.8 27B on a Mac Studio M4 Max. It was slow and generated R code using yahoofinancials—a Python package. Lower-tier hosted models like GPT-5.6 Luna are faster, cheaper, and more accurate. At $0.20 per M tokens, a $2,000 Mac buys roughly 10 billion tokens. Local models offer privacy. But once you count hardware, time, electricity, and debugging, local inference may be a false economy. Hosted inference may also have a smaller environmental footprint per token, thanks to specialized hardware and better utilization. I wrote this to resist the temptation of the new Mac Studio M5 that will be released next week. #LocalAI #LLMs #AI

Where are your differentially expressed genes on the genome? Our tool (ShinyGO ge-lab.org/go/) was used to discover that Chr. 1q21.3 amplified, triple-negative breast tumors are more responsive to common cytotoxic therapies. doi.org/10.1101/2020.04.29.0…

Live voice in Codex has completely changed how I use AI for academic research. In this video, I’m speaking naturally with Codex while it opens notes from my Obsidian vault, searches for papers in Zotero, brings the evidence together, answers my questions, and creates a new linked note all by voice alone. I can ask it to search and find articles in my reference manager as well as find and write into my notes in Obsidian. After two weeks, I’ve almost stopped typing. I can continue the conversation from my phone, wherever I am even while driving long distances. It feels less like operating software and more like discussing a problem with a knowledgeable senior colleague who understands my research. I’m doing a PhD, and this is already transforming how I explore ideas, interrogate evidence, and build knowledge. And I’ve only just begun. I’m amazed more people aren’t talking about this. Live voice in Codex is extraordinary.

Codex beats Claude Code in a bioinformatics benchmark: 93.8%: Codex + GPT-5.6-sol-xhigh 92%: Codex + GPT-5.5 91%: Claude Code + Opus 4.8 Ultra CompBioBench v1 (Surag Nair @suragnair) huggingface.co/spaces/Genent…

Seeing a similar trend on our benchmark: huggingface.co/spaces/Genent…

For bioinformatics coding, use Codex with gpt-5.6-sol. Beats Claude Code with Opus 4.8 by a long shot. Fable 5 won't even look at anything remotely related to biology.

“Explain briefly in plain language, using bullet points.” I’ve been saying this to LLMs a lot recently. These models have become incredibly powerful. But their answers are often long, technical, and overly detailed. Sometimes I simply don’t have the time, patience, or knowledge to read through everything.

My 5 strategies for saving tokens—especially with Claude Fable 5: - Use Claude.ai for bounded tasks. Use Claude Code only when it's complex and involves editing many files. - Control the context. Start a new chat when switching tasks. Edit a prompt instead of piling on. Regenerate responses strategically. DON'T keep on chatting!!! - Minimize output tokens—they cost 5× more. Ask for only the diff or revised section, not the entire script or document. - Ask for brevity. I often add “Be brief” or “Use bullets” to my prompts. Also avoids reading page-long responses. - Lower the effort setting. Use maximum effort only when it requires deep reasoning. Powerful models are naturally thorough. Make them earn every token.

Fable 5 extended again to 7/19. Thank you, Anthropic, for motivating me! I needed deadlines, especially in the summer!

I still upload key files to the Claude or ChatGPT websites to get a fresh perspective. Claude Code or Codex can be influenced by what's in the local project folder. CLAUDE.md, AGENTS.md, and project memory files are helpful but could also prime AI into certain directions. Coding agents have become so complex. These web interface provides a fresh, controlled context.

Fable 5 is available now! It's Anthropic's most powerful model, previously banned by the US government.

Don't worry about not keeping up with AI. Nobody is. Identify and solve a problem! Even last year's AI is underutilized! For scientists, discover something new. Or create a tool that helps. We are licensing standalone, customizable copies of iDEP. bioinformatics.sdstate.edu/i…