Reddit files a lawsuit against Perplexity and three data scraping companies, accusing them of illegally stealing its data by scraping Google search results
comparing them to “bank robbers” and accusing them of sidestepping its controls to scrape its data and get rich off the AI gravy train. https://www.businessinsider.com/ ... Andrew Curran / @andrewcurr...
Gumroad founder Sahil Lavingia discusses working as a DOGE software engineer for the VA and says he was let go after his Fast Company interview was published
Gov't tech workers I spoke to went bonkers when DOGE workers began connecting non-government laptops to their systems. … Jacob Harris / @jacobharr.is : Reading Sahil Lavingia's blog post about working...
Six AI and ML experts detail what they learned from building real-world applications on top of LLMs over the past year, including common prompting pitfalls
‘What We Learned from a Year of Building with LLMs (Part I)’ — https://www.oreilly.com/... Hallucinations are still a big problem; RAGs are better than fine-tuning; long context models don't help m...
The US IRS starts testing its free, “mobile friendly” English and Spanish direct e-filing system, with some low- and moderate-income taxpayers in 12 US states
> The direct file pilot will be open to low- and moderate-income taxpayers with simple returns. … X: Daniel Castro / @castrotech : “...the I.R.S. could fill out tax returns with information it already...
Facebook's wav2vec Unsupervised, a way to build speech recognition systems that require no transcribed data, may bring automatic translations to more countries
Wav2vec Unsupervised (wav2vec-U) … Tiernan Ray / ZDNet : Facebook AI cuts by more than half the error rate of unsupervised speech recognition Donald Conway / Insider Voice : Facebook Wav2vec-U learns ...