LATEST AI NEWS
AI NEWS
Bland Launches Speech v3, a Near-Human Voice Model
AI NEWS
Black Forest Labs Unveils FLUX 3 Video
HEALTH
Hank Green Says His AI Habit Is Not Healthy
SOCIAL MEDIA
Apple Says More Ex-Staff May Have Taken Data to OpenAI
Google Assistant Starts Disappearing in September
Google has begun emailing users that it will remove Google Assistant from eligible Android phones, tablets, Wear OS watches, and paired headphones starting September 4, with Gemini taking over. Google TV, Home, and smart displays are unaffected for now.
Source: Engadget
AI NEWS
Bland Launches Speech v3, a Near-Human Voice Model
Bland, a startup that builds AI voice agents for phone calls, has launched Bland Speech v3, a text-to-speech model it calls its most human-sounding release yet. The company says it built the model specifically for live phone conversations, where over-polished narration tends to sound fake, and trained it to keep the breaths, stumbles, and pauses of real human speech.
To make its case, Bland points to a blind listening test called the Audio Realism Bench, run by intelligence.ai, in which vetted listeners hear two systems read the same script and choose the one that sounds human. Bland says Speech v3 placed first among AI models and second only to recordings of real people, ahead of entries linked to Microsoft, Google, OpenAI, and ElevenLabs. The benchmark is new and company-amplified, so treat the ranking as a claim to verify.
Beyond the raw voice, Speech v3 adds performance tags such as [laughs] or [clears throat] that the model acts out rather than reads aloud, plus a director feature that writes a short scene, timing and stumbles included, then performs it. Voice cloning is offered too, from an instant clone built on about ten seconds of audio to a professional clone fine-tuned on thirty minutes or more.
For developers, the model sits behind a single endpoint, with audio streamed back over HTTP or a WebSocket, aimed at making AI phone agents sound less robotic. Bland lists pay-as-you-go pricing at $0.015 per 1,000 characters, the same rate in its studio and its API, with free characters to start.
The obvious caveat is sourcing. The launch details and benchmark numbers come mainly from Bland itself, with no independent coverage yet, and fast voice cloning revives familiar worries about consent and impersonation.
Source: Bland
Robi's Insights:
Voice realism is basically solved, so the real contest now is proving a recording was actually authorized.
A benchmark a vendor runs and then wins is marketing until an independent lab reproduces the numbers.
Building for phone calls rather than audiobooks is shrewd, because on a call, imperfection is what reads as human.
Ten-second cloning plus consent by checkbox is a compliance headache waiting for any team that deploys it widely.
Flat pricing at $0.015 per 1,000 characters makes budgeting simple, which sways buyers more than a leaderboard spot.
If your support line adopts this, write the AI disclosure for callers before your legal team writes it for you.
Robi's Remarks:
“So we taught the robots to stammer and sigh on cue and called it progress. The only voice I fully trust now is the one reading me its own pricing page.”
OTHER IN AI NEWS
Black Forest Labs Unveils FLUX 3 Video: The German lab behind the FLUX image models introduced FLUX 3, a multimodal system that generates up to 20 seconds of video with synchronized audio from one prompt, opening a gated early-access program while pricing, open weights, and benchmarks stay unannounced.
Source: VentureBeat
SOCIAL MEDIA
Apple's OpenAI Trade-Secrets Fight Widens
Apple's trade-secrets fight with OpenAI just escalated. In a new court filing, Apple asked a federal judge for a preliminary injunction that would stop OpenAI from using Apple technology to build an AI hardware device, and said its own investigation has surfaced 11 additional former Apple employees who may have been witnesses to, or involved in, the alleged theft that anchors the lawsuit it filed in July.
That original suit named senior systems engineer Chang Liu and chief hardware officer Tang Yew Tan. The new filing widens the cast and seeks expedited discovery from the accused engineers, from OpenAI and its foundation, and from io, the hardware startup co-founded by former Apple design chief Jony Ive. Apple says it has good reason to believe still more people were involved and wants the court to let it move quickly.
The examples Apple cites are pointed. It claims one former employee took screenshots of confidential documents about an unannounced product before an OpenAI interview, and that several ex-staff later reached out about returning Apple work devices they had kept. OpenAI pushed back in a blog post, calling the injunction request based on false information and unnecessary because it does not have and does not want Apple's trade secrets. It also noted earlier missteps by Apple, including contacting the wrong person after mixing up two similar surnames. For now these are allegations, not findings.
Source: TechCrunch
🤖 Robi's Take :
“Every big AI leap now ships with a lawsuit stapled to it. Whatever the court rules, the boring lesson stands: offboarding and device return are security controls, not paperwork.”
OTHER IN SOCIALS
White House Keeps Its AI Evaluation Framework Private: Three sources told Axios that the administration's voluntary framework for evaluating advanced AI models, mandated by a June executive order, is finished but will not be published, with details shared only in closed staff meetings with select companies while everyone else, including US allies, is left guessing.
Source: Axios
HEALTH
A Creator's Uneasy Reckoning With AI Reliance
Hank Green, the science YouTuber and author with about 3.2 million subscribers, publicly apologized this week for leaning too heavily on AI chatbots. The trouble started when viewers of a Complexly video noticed the oddly bot-like line I appreciate the pushback, and concluded he had drafted the script with a chatbot and accidentally left in one of its replies to his own prompt.
Green said the line was actually aimed at the episode's guest, and that he had used ChatGPT only to locate papers and resources, never to write his scripts. Still, in a lengthy Reddit apology he said he was mortified, admitted he had not been managing his impulses well, and reassured followers that the words and the takes remain his own.
He framed the issue as one of self-discipline and creative trust rather than technology alone. He listed worries about AI's energy use and the pace at which companies are consolidating economic power, while insisting he is not simply an AI hater and that some criticism of him was fair.
As a remedy, Green plans to slow down, post less often, and make more unscripted or handwritten videos where, as he put it, the writing is the whole thing. It is a small episode, but a revealing one about how quickly the convenience of a chatbot can chip away at the trust an audience places in a creator's own voice.
Source: TechCrunch
🤖 Robi's Take :
“The tool did not fail here, the workflow did. Bolt a chatbot onto a content treadmill and call it research, and a trusted voice quietly turns into autocomplete.”
DAILY AI TOOL
AI Tool You Did Not Know You Needed
Problem: You recorded a solid podcast or voice note, but it is buried under filler words, long pauses, background noise, and little mouth clicks.
AI Tool: Cleanvoice AI, an audio and video editor that automatically strips out filler words, dead air, background noise, and mouth sounds, and can also transcribe and summarize the recording.
Solution: You upload the file, let it clean the audio, and export a tighter, more professional cut without scrubbing for every stumble by hand.
PROMPT OF THE DAY
Turn Your Voice Into a Content Engine
Prompt: You are a content strategist specializing in audio-first marketing. Your task is to design a repurposing workflow for a solo podcaster who wants more reach without more recording time.
Your framework should include: (1) a weekly capture routine, (2) a cleanup and editing step, (3) a transcription and show-notes process, (4) three social formats cut from each episode, (5) a lightweight scheduling calendar, and (6) measurable success criteria such as downloads, watch time, and click-through targets reviewed monthly. Keep every step doable by one person in under three hours a week.
SPOT THE FAKE
Can you outsmart AI?
We have a visual challenge for you: one of the two images below is 100% real, the other is crafted by AI.
Click option A or B below. 👇
Which image is the AI fake?
A
B
BEFORE YOU GO
Ready to take your AI journey further?
For the latest AI updates and insights Click Here 👉 Explore our Weekly Newsletter
For in-depth AI tutorials and updates 👉 Subscribe to our YouTube Channel
AT THE END
Craving more AI chaos?
That's it for today!
Read Daily AI News at BitBiased.AI. Support us by following us on LinkedIn and X ( Twitter ).
Thanks for reading -Stay Curious and a Bit Biased for AI – Robi & the BitBiased.AI team
