Highlights
What happened at the companies we cover, by day, newest first. Every item links to its sources.
Showing 1–20 of 60 highlights
Tuesday, Oct 20 60 stories
Naver discloses an investment in Suno as it courts US AI startups
NewsNewsXBluesky
Korea's Naver has invested in Suno, according to KED Global and Korea Economic Daily, as part of a push into AI services beyond its own language models. The reports say the Suno stake came via Naver's US venture arm through a fund, alongside a separate bet on AI assistant startup Instinct; timing and size of the Suno investment were not disclosed. Separately, social chatter about Suno on X and Bluesky was dominated by users sharing songs, with a vocal minority of musicians criticizing AI music and some users complaining that the V6 model is worse.
Personal Agent Protocol from Sierra and Meta keeps drawing commentary
DiscussionXNews
Posts on X and a few news roundups kept discussing the Personal Agent Protocol, the draft open standard Sierra and Meta published this month to govern how personal AI agents identify themselves to businesses and act for users. Commentators linked it to recent friction such as retailers blocking shopping agents, and said Walmart, Shopify and Stripe are among those involved. Advice aimed at customer experience and e-commerce teams focused on tracking agent contacts separately and preparing listings to be read by agents. The protocol is still early, and posts treated it as a direction of travel rather than a finished standard.
Sources
Sierra and Stripe let AI agents take card payments during phone calls
LaunchNewsX
Sierra announced a Stripe partnership that lets its customer service agents collect card payments in the middle of a call. The agent switches to a secure mode, the caller types their card number on the keypad, and the details go straight to Stripe, so the agent sees only the payment result. Healthcare billing company Cedar is an early user and said 71.3% of calls that start a payment with its agent end in a successful payment. Sierra is also now in the Stripe App Marketplace, and the two companies are working together on the Personal Agent Protocol. The Next Web, Unite.AI and others covered it, and replies to Sierra's post on X were positive about the conversion figure.
Sources
Together AI cuts B200 post-training price to $9 per GPU-hour for eight weeks
LaunchX
Together AI announced on X that post-training jobs on NVIDIA B200 GPUs will cost $9 per GPU-hour for the first eight weeks, down from $11. The company pitched the service for running supervised fine-tuning or reinforcement learning while keeping the training loop in customers' own code. The same day it hosted an open-models-in-production meetup in Berlin with engineers from n8n and NVIDIA around GTC Berlin and promoted an upcoming launch event with StepFun in San Francisco. Replies and reposts were few but positive, with some framing Together as the go-to vendor for teams wanting to own their models.
Sources
Replit prepares Japan launch with Tokyo hires and community events
HiringX
Replit is building out a presence in Japan. The company is hiring a Tokyo-based growth lead and account executive, and has scheduled an invite-only Japan opening night in early November. On X, Replit staff and community members celebrated what they described as Replit's first office in Asia and promoted a meetup with a well-known local developer. Reaction was small but enthusiastic, mostly congratulations from community builders.
Sources
Debate continues over Replit switching its agent default from Claude to OpenAI
DiscussionXBluesky
Commentary kept building on Replit's decision to make OpenAI the default model for Replit Agent, after CEO Amjad Masad said on stage that the company was no longer working with Anthropic and then clarified that Replit stays model-agnostic and Claude is still selectable. TechCrunch reported the Anthropic partnership is still in place. Posts on X and Bluesky argued that because most users never change the default, the switch matters a lot, and some framed it as a lesson in AI vendor risk. A few posts were openly skeptical of how the announcement was handled.
Sources
Higgsfield creators flood X with AI films and trending templates
PraiseXBluesky
Creators continued to post short films, ads and template videos made with Higgsfield, many tagged as partner content. The most engaged was a stylized historical short that drew roughly 500 likes and reposts, and a football-goal template was said to be trending on Higgsfield's main page. Others showcased portfolios built from Higgsfield films, motion-control tests and the Katana editing feature. Sentiment was strongly positive, though one user complained about being charged a renewal after a looping cancellation screen.
Sources
Director laments AI copies of Yung Lean 'Storm' video, many made with Genjutsu
DiscussionNews
Stereogum, summarizing a Vulture interview, reported that director Romain Gavras is dismayed by a wave of AI-generated imitations of the GENER8ION and Yung Lean 'Storm' music video. Vulture noted that many of the copies, which swapped the dancers for dogs, celebrities and controversial figures, were made with Higgsfield's Genjutsu tool. Gavras said he turned down AI companies that offered to license the video and argued the copies lack the human quality of the original. The piece frames Higgsfield's tool as a driver of low-effort trend content, a criticism that contrasts with creators' enthusiasm on X.
Sources
Higgsfield launches Genjutsu Lite, a cheaper way to put yourself in trending videos
LaunchX
Higgsfield released Genjutsu Lite, a faster and cheaper tier of its video tool that keeps a clip's motion and camera work while swapping in a user's own face, characters or objects. The company says generations cost about 3.5 times less at 720p and 5 times less at 1080p, with clips up to 15 seconds, and the feature is available in its app and via API, free to try in some regions. Its launch post drew several hundred likes and reposts, and creators quickly shared results. Commentators highlighted that lower prices make large-scale testing of such swaps more practical.
Sources
Status trackers log repeated Fireworks degradations on popular open models
ComplaintBlueskyX
Automated status trackers on Bluesky logged several Fireworks AI incidents over the day, with degraded performance on GLM 5.3, GLM 5.3 Flash, DeepSeek V4.1 Flash and Kimi K3. Each incident was later marked resolved, and an X status digest also listed Fireworks among services with incidents on October 20. There was little user commentary beyond the alerts themselves. The pattern adds to reliability questions as the company courts investors at a much higher valuation.
Sources
Fireworks AI reportedly in early talks to raise at a $30B to $40B valuation
NewsNewsXBluesky
Business Insider reported, citing people familiar with the matter, that Fireworks AI is in early talks for a new round valuing it at $30 billion to $40 billion, roughly double the $17.5 billion it fetched in July. The article notes the company said in July it had passed $1 billion in annualized revenue, and that the talks come as it deals with a recent security incident in which an intruder used internal credentials to access some customer data. Fireworks declined to comment. The scoop spread quickly on X and Bluesky in several languages, with investors debating its GPU rental economics, while a separate tracker said the company grew headcount 56% in 90 days.
Sources
Profound adds Moz SEO data and partners with Bazaarvoice on AI search visibility
LaunchXNews
Profound announced two data partnerships on the same day. Moz metrics such as Domain Authority now show inside Profound's Pages feature next to its answer-engine visibility data, so its AI marketing agent can act on both. Separately, Bazaarvoice and Profound said they will help brands measure how they appear in AI search and improve it by publishing reviews and other user content in machine-readable form. Bazaarvoice claimed a median 40% lift in AI-driven referral traffic for its package. Reaction on X was small but positive, with users asking whether the new data will also be available through Profound's MCP server.
Sources
Bluesky users slam a game developer for quietly using ElevenLabs voices
ComplaintBluesky
A wave of Bluesky posts criticized an indie game developer who, after initially claiming no AI-generated assets, acknowledged replacing a human voice actor with ElevenLabs voices and using other AI tools. Critics argued the disclosure came too late and said creators should avoid paying ElevenLabs and Suno at all, with some calling ElevenLabs one of the shadiest AI options and accusing it of using voices without consent. A few replies tried to separate older text-to-speech tools from cloning-based ones. The thread drew the most negative engagement about ElevenLabs on Bluesky today, with little defense of the company.
Sources
ElevenLabs launches Eleven Realtime speech-to-speech models for voice agents
LaunchXNewsBluesky
ElevenLabs released Eleven Realtime, a pair of speech-to-speech models that bundle recognition, turn-taking, reasoning and voice output into one system for voice agents. The company says the Max version answers 98.8% of Big Bench Audio questions correctly while starting to speak in about 0.57 seconds, and the cheaper Mini can tell a finished sentence from a listener's brief acknowledgment. Both are available in ElevenAgents and through an OpenAI-compatible Realtime API, priced per minute plus tokens. The launch post was the company's most engaged of the day, with early testers, including a Japanese developer, praising natural conversation quality, while The New Stack noted developers give up model choice for the simplicity.
Sources
Tiny says its 2023 Poolside seed check is now worth about 44x on paper
NewsXNews
Tiny Ltd., the Canadian holding company chaired by Andrew Wilkinson, disclosed that its $250,000 seed investment in Poolside now has an implied value of roughly $11 million. The figure is based on the reported terms of Nvidia's August deal, an equity investment plus a large license for Poolside's Model Factory. Kalkine Media noted the mark is unrealized, depends on terms Tiny has not verified, and gives no value to Tiny's separate stake in the Poolside Infrastructure spin-out building Project Horizon in West Texas. Wilkinson's post on X drew over 100 likes and a wave of congratulatory replies praising founder Jason Warner, along with a few jokes about paper gains.
Sources
Runway Creative is now sold through Anthropic's Claude Marketplace
LaunchNews
Runway announced that its Runway Creative platform for generating and editing video, images and audio is now listed in Anthropic's Claude Marketplace. Eligible enterprise customers can apply part of their existing Anthropic spending commitment to Runway purchases. Runway said Claude powers the reasoning in its Runway Agent, which turns a brief into a shot plan and assembles finished videos using Runway and third-party models. Coverage was limited to the company's announcement and a few reposts.
Sources
Runway CEO says AI could replace human actors, stirring VFX debate
DiscussionNewsBlueskyX
At the View Conference in Turin, Runway co-CEO Cristóbal Valenzuela answered an audience question by saying AI could definitely replace human actors and that this is already starting to happen. Variety reported the remark sparked one of the event's biggest debates, with an NVIDIA executive on stage arguing that actor-director spontaneity can't be prompted. Valenzuela also pushed back on the idea that models are databases of training material, while Variety noted Runway is a defendant in a visual artists' copyright class action. The quote was syndicated by IMDb and others and shared on Bluesky and X, drawing mostly skeptical or critical replies.
Sources
Debate over OpenEvidence's JAMA usage paper continues
DiscussionNewsXBluesky
Discussion of the JAMA paper describing about 271 million clinical questions from roughly 952,000 US clinicians on OpenEvidence carried into a second day. STAT's health tech newsletter argued the paper leaves important questions open, while medical accounts on X summarized the usage figures, noting treatment decisions were the top query type. Doctors on X were split between enthusiastic regular users and critics who view the product as a pharma advertising channel, a theme also raised in an eMarketer piece on physician search ads. A Financial Times feature on the company's rapid rise among US doctors added to the attention.
Sources
OpenEvidence and Doximity settle their bitter clinical AI lawsuit
NewsNewsX
OpenEvidence and rival Doximity have settled the dueling federal lawsuits that began when OpenEvidence accused Doximity staff of posing as doctors to probe its AI for proprietary details. A Massachusetts judge dismissed all claims and counterclaims with prejudice, each side pays its own legal fees, and no other terms were disclosed. Healthcare Dive noted the deal came as OpenEvidence faced a deadline over allegedly deleted internal Slack messages, and Fierce Healthcare and Law360 also covered it. Reaction on X was mostly factual sharing, with a few posts taking jabs at both companies.
Sources
Perplexity's SPACE agent sandbox joins Crusoe's managed AI lineup
NewsX
Perplexity CEO Aravind Srinivas said the company's newly open-sourced SPACE sandbox platform for AI agents will become part of Crusoe's Managed AI products, calling Crusoe well placed to become a full-stack cloud spanning power, compute and services. The move extends the multi-year Crusoe and Perplexity partnership announced in September. Most reaction treated it as validation of Crusoe's push up the stack, while one skeptical reply argued full-stack branding only matters if it lowers cost per token. Commentators also cited Crusoe among neoclouds now buying or adding software rather than just renting GPUs.
Sources