X open-sources its ranking algorithm, adds shadowban checker
The black box opens a crack. X released the source code behind its "For You" feed and shipped a tool that tells users when the algorithm has quietly demoted them — while AMD and Suno made moves of their own in the AI buildout.
X open-sourced the code behind its "For You" feed and core ranking engine, and added a transparency tool that lets users check whether the algorithm has shadowbanned them. The announcement Thursday puts the feed's model configuration, filters, and ranking parameters — the weights that decide which posts surface — on GitHub under an Apache v2 license, making the codebase roughly 10 to 15 times larger than anything X had previously published. Users who posted ten or more times in the past month can download their account's aggregate stats from a new "Under the Hood" page in settings and see which labels were applied to their account or posts; X notes that non-technical users can hand that file to an LLM for an interpretation.
X's pitch is that the code is auditable: external researchers have already run the per-post "score" outside the company, and VP of product Keith Coleman says engineers will consider community pull requests. The Grok-powered systems that predict rule-violating posts stay closed, to keep bad actors from gaming moderation. After years of shadowban accusations — leveled at the platform since long before Musk owned it — this is the most concrete transparency commitment X has made. It doesn't make the company more open overall (it has become less forthcoming on user metrics and government takedown requests since going private), but it turns the algorithm itself into a public artifact. The open question is whether that artifact stays trustworthy as the shipped code and the deployed system drift apart.
AMD is planning to raise as much as $5 billion in what could be its biggest-ever investment-grade bond sale, funding a spending ramp-up driven by AI demand. Bloomberg reported the chipmaker is tapping the debt market as it pours money into data center silicon, the latest sign that the AI infrastructure bill is being financed with borrowed money as much as equity — we covered the stock-side precedent last month in Intel raises $15B in stock offering to fund AI buildout. Debt without dilution is becoming the default for AI capex, and AMD stepping up at this scale shows how much the race for AI compute now costs.
Suno Studio 2.0 turns the AI music app into something closer to a real production tool, adding MIDI, effects plugins, and a chatbot that helps you produce. MIDI — the feature Suno says users requested most — lets you play in chords and use them as a prompt: extend a melody, generate a B-part, or record and edit notes directly on the timeline, with a built-in wavetable synth standing in for third-party plugins. The new chat assistant understands the session's musical context, so you can ask it to write a guitar line, tidy up the arrangement, or make a vocal track "sound better" — it layers on reverb, compression, or EQ, and can even design custom effects plugins, free of credits for now. Studio 2.0 also adds automation curves and stem separation, with multitrack exports for Premier subscribers. It's a deliberate pivot from novelty generator to professional DAW — the company behind 100 million users (Suno hits 100M users as UMG, Sony seek $9B) wants to live inside the working musician's workflow, not just the meme pipeline. Whether musicians accept an assistant that does half the engineering is the open question.
What to watch: whether X actually merges community pull requests into the production ranker — and who ends up auditing the auditors.
Would you trust an open-source feed algorithm over a closed one? Tell us in the comments.
