2026 · 05 · 06 · Wed

Updates for 5/6

ChatGPT reveals which memories shaped each reply, Gemini 3.1 runs the smart home, and ElevenLabs hit an $11B valuation. A $200K prompt injection shows the risk.

A · Theme of the day

AI is learning your life and running your home

ChatGPT shows the memories behind each reply; Gemini 3.1 runs the whole home.

ChatGPT now shows which memories shaped each response

ChatGPT (OpenAI)

Compared to before

Until now users could only view a list of saved memories. The new panel surfaces the exact memories behind each answer, making personalization transparent for the first time.

What changed

New "memory sources" feature shows which memories influenced each response; past-chat, file, and Gmail personalization rolling out to Plus/Pro (web) first.

Why it matters

Users can audit whether unintended personal details shape replies, or build a memory bank to skip re-explaining context. Rollout starts with Plus/Pro (web).

Source: chat.openai.com　·　→ View on chaos map

Google Home now handles multi-step commands in plain language

Gemini (Google)

Compared to before

Smart home assistants worked one command per action, so multi-device routines needed manual setup. Gemini 3.1 parses multi-step sequences from one sentence.

What changed

Google Home updated to Gemini 3.1 — complex multi-step smart home tasks now run from a single natural-language command.

Why it matters

A movie night or morning routine — lights, temperature, blinds — now takes one sentence. Households deep in the Google ecosystem benefit most.

Source: gemini.google.com　·　→ View on chaos map

B · Theme of the day

Voice AI's $11B round proves it is no longer experimental

ElevenLabs closed a $500M+ Series D at an $11B valuation on $500M ARR.

ElevenLabs closes $500M+ Series D at $11B valuation

ElevenLabs

Compared to before

Known for expressive TTS and voice cloning in 32+ languages, ElevenLabs expanded from content creation into enterprise dubbing and call center agents, with ARR accelerating.

What changed

ARR exceeded $500M (about 43% YoY); $11B valuation, Series D over $500M with BlackRock, Nvidia, Jamie Foxx and others.

Why it matters

$500M+ revenue growing 43% a year signals voice AI is core infrastructure. Nvidia's participation suggests GPU-level integration and lowers counterparty risk for API users.

Source: elevenlabs.io　·　→ View on chaos map

C · Theme of the day

Faster local AI and a $200K prompt-injection loss arrive together

Local coding inference doubles in speed; a $200K prompt-injection loss lands.

Gemma 4 MTP doubles coding inference speed on Mac

Gemini / Google DeepMind

Compared to before

Gemma is Google's open-weight family for local hardware. Earlier versions generated one token at a time; MTP predicts several at once.

What changed

Gemma 4 MTP (Multi-Token Prediction) released — 2x+ inference speedup for coding tasks on Mac via speculative decoding.

Why it matters

Mac developers get faster local coding AI without cloud APIs, making open-weight inference more competitive for completions and quick scripts.

Source: deepmind.google　·　→ View on chaos map

Morse code prompt injection drained $200K from a crypto bot

Grok (xAI)

Compared to before

Prompt injection has long been a warned-about risk for AI agents, but real financial losses were rare outside research settings.

What changed

Prompt injection via Morse code tricked a crypto bot into transferring $200K — the risk of executing LLM output without validation.

Why it matters

Any pipeline passing AI output straight to an action system is vulnerable. High-stakes operations need a validation layer or human confirmation.

Source: grok.com　·　→ View on chaos map

Past updates

A daily archive of changes actually applied to the site.