In partnership with
FROM OUR PARTNER
100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic and OpenAI say AI now writes 100% of their code.
If you're not using AI, you're spending 40 hours doing what they do in 4.
These 100+ Claude Code hacks fix that and help you ship 10x faster.
Sign up for The Code and get:
100+ Claude Code hacks used by top engineers — free
The Code newsletter — learn the latest AI tools, tips, and skills to code faster with AI in 5 minutes a day
TODAY AI BRIEF
3 things to know about Block's new voice model
1 Block unveiled Dessa, a voice model trained on conversation alone.
Block's research lab announced Dessa on October 23 as a research preview. Block says it was trained from scratch on recordings of human conversation, with no text and no pretrained model. Block also claims it learned from hundreds of times fewer words than a text language model needs. Read the launch coverage.
2 It listens and talks at the same time, which is harder than it sounds.
Dessa is a full-duplex model, meaning it does not wait for you to finish your turn before it responds. Block says it learned from real conversations when to interrupt and when to give a short "yeah." Those timing details are Block's own description. Read RuntimeWire's report.
3 You cannot use it yet, and key specs are missing.
A public waitlist is open, and early access goes to a small group of AI safety researchers. No model weights, API, license or pricing has been announced, and there is no parameter count or total audio hours. Block says it runs on a MacBook Pro with an M4 Max chip or better, faster than real time, but it has not published the figures behind that claim. Read the coverage.
AI FORWORD
Voice AI is learning the rhythm of a conversation, not only the words
The usual approach. Most speech-to-speech systems start from a text language model and then add a voice on top. They tend to wait for you to finish, then reply. Block's approach differs, according to the coverage, because it starts from audio.
The claim. Block says Dessa picks up interruptions, short acknowledgments and even laughter and sarcasm from real talk. Those are the things speech-to-text pipelines usually lose. RuntimeWire's account describes the release, but the figures are Block's own.
The gap. There is no paper, no benchmark and no independent test yet. Launch demos reportedly show the model laughing mid-sentence and finishing a user's sentence, but the coverage does not verify them. Check the demo notes before you repeat them.
The reframe. Judge a voice assistant by its timing as much as its voice. A pleasant voice that cuts you off is worse than a plain one that waits its turn, and this release makes timing the thing to test.
PRESENTED BY :
9 Plays. One Mission: Keep More of Your Holiday Profit.
Holiday profit rarely disappears all at once. Deeper discounts, rising ad costs, fulfillment fees, and returns take their share little by little.
Get 9 Plays to Defend Your Margin This Holiday Season, tactical advice from ecommerce experts to uncover where profit gets lost and what you can do to keep more of it this Q4.
SOCIAL SNAPSHOT
What to check before you believe the launch
Trained without text. Block says Dessa has no text training and no pretrained model. That is a company claim and has not been independently checked. Read the claim in context.
Laptop claim. Block says it runs on a MacBook Pro with an M4 Max chip or better, and that training used a single NVIDIA GB200 NVL72 rack. Neither figure is verified. See the hardware notes.
Early access is narrow. Early access goes to a small group of AI safety researchers, and a public waitlist is open. No general release date has been given. Read the access details.
Who built it. Jack Dorsey introduced Dessa, and the project is led by Ragavan Thurairatnam, who co-founded the earlier Toronto startup Dessa. Read the team details.
PROMPT STATION
Test any voice assistant's timing before you trust it
A voice assistant can sound natural and still talk over people. This prompt works in Claude, ChatGPT or Gemini and writes a short test script that checks interruptions, pauses and acknowledgments. Run it on any voice tool before you put it in front of a customer.
I am testing a voice assistant for [THE BUSINESS TYPE, SUCH AS A CLINIC OR REPAIR SHOP]. Its main job is [THE CALL TASK, SUCH AS BOOKING AN APPOINTMENT].
Write 10 short test scripts I can read aloud to it. Include three where I interrupt it mid-sentence, three where I pause for two seconds before finishing, two where I say only "yeah" or "mm-hm", and two where I change my mind halfway through.
For each script, tell me what a good response sounds like and what a failure looks like, such as talking over me or ignoring the acknowledgment.Tip: Swap the brackets for your real use, such as a dental clinic that books cleanings or a phone repair shop that quotes screen fixes. For a sharper test, record each run and count how many times the assistant talks over you.