E-Safety Commissioner reckons she could do with some new regulatory powers

E-Safety Commissioner Julie Inman Grant fronted up to a senate hearing on proposed law changes to criminalise sexual deepfakes. She talked about "de-clothing", "nudification", "tracking and monitoring apps", "the mass creation of fake and imposter accounts" and social media networks "allowing these predators and criminal organisations to literally colonise their platforms and target people on the platforms". No mention of what they would do to try and solve all those things we agree are bad and suck, except "now is the time to be thinking about the kinds of powers that we might need to make us more effective at a systemic level". What would those powers be? Doesn't say, but there is "a range of tools I think we need" and that she'd "welcome that kind of power" to "de-platform" apps and accounts when tech companies don't do it.

Meta's new Llama 3.1 405B LLM is open source and as good as GPT-4o

Meta unleashed Llama 3.1 405B, an open-source LLM that's as good as other GPT-4-ish LLMs like OpenAI's GPT-4o, Claude's 3.5 Sonnet and Google Gemini 1.5 Pro. Meta apparently "trained the 405B model on over 15 trillion tokens of training data scraped from the web (then parsed, filtered, and annotated by Llama 2), using more than 16,000 H100 GPUs". The reason you'd care about Llama 3.1 405B is that you can run in on your own hardware, unlike all the other competitors in the same GPT-4 class, who require you to pay for access to an API. This means you could see other companies take this, run it on their own infra and use it in places where paying Google or OpenAI to use their API for each request is cost prohibitive, slow or a privacy risk. Zuck reckons "the Llama 3.1 release will be an inflection point in the industry where most developers begin to primarily use open source, and I expect that approach to only grow from here".

Crowdstrikes squeezes out a bit more info on what happened last week

Crowdstrike has released a Preliminary Post Incident Review today, giving us the most detail yet on what the fuck happened last Friday arvo. There's a lot of Crowdstrike product specific terminology, but "due to a bug in the Content Validator, one of the two Template Instances passed validation despite containing problematic content data. Based on the testing performed before the initial deployment of the Template Type (on March 05, 2024), trust in the checks performed in the Content Validator, and previous successful IPC Template Instance deployments, these instances were deployed into production". A dud file somehow passed their validation procedures, which because it passed, didn't go into any further testing as it was based on a template from March that did work. Still waiting on a root cause analysis, but Crowdstrike reckons that'll come soon and will be public.

The Swiss government will make all code it pays to develop, open source

Switzerland recently passed a law called the Use of Electronic Means for the Performance of Official Duties (EMBAG) that mandates "the use of Open Source Software - to the extent possible, the federal government should disclose the source code of software that it develops or has developed for free reuse" and "the gradual public provision of data obtained or generated for the fulfillment of statutory tasks (Open Government Data)". Makes a lot of sense to me that if public funds are used to develop software, that software should be made available publicly too.