LATEST AI NEWS
AI NEWS
DeepSeek Launches V4-Flash Coding Agent
AI NEWS
LinkedIn Adds an AI-Slop Report Button
HEALTH
Brain-Inspired AI Plans on Far Less Energy
SOCIAL MEDIA
OpenAI Gives 100,000 Academics Free ChatGPT
HeyGen Turns Any Doc Into a Video Podcast
HeyGen’s new Video Podcast tool spins a document, link, or idea into a two-host video show, studio scenes, multi-camera cuts, and B-roll included in minutes. It drew instant buzz and equally instant “why watch AI hosts?” pushback online.
Source: Digg
AI NEWS
DeepSeek Launches V4-Flash, a Cheaper Coding-Agent Model
DeepSeek V4-Flash entered public beta on July 31 through the company’s official API, and the story is efficiency: the lightweight model now beats DeepSeek’s own pricier V4-Pro-Preview on several coding-agent benchmarks. The company says the leap came entirely from post-training reinforcement learning and distillation with no change to the model’s architecture or size.
Under the hood, V4-Flash is a Mixture-of-Experts model with 284 billion total parameters but only 13 billion active per query far smaller than V4-Pro’s 1.6 trillion total and 49 billion active. It pairs two compression techniques, Compressed Sparse Attention and Heavily Compressed Attention, to keep long-context work cheap.
The upgrade targets autonomous “agent” work: executing multi-step tasks and calling tools. DeepSeek reports scores of 82.7 on Terminal Bench 2.1, 54.4 on DeepSWE, and 76.7 on Cybergym everal above the larger Pro-Preview. V4-Flash natively speaks OpenAI’s Responses API format and is compatible with Codex, so existing agent tooling can plug in.
Price is the pitch. V4-Flash runs about $0.14 per million input tokens and $0.28 per million output, against $0.435 and $0.87 for V4-Pro, with a 2,500-request concurrency ceiling aimed at high-volume automation.
For now only the API is upgraded; DeepSeek’s app and website are unchanged, and the full V4-Pro release is still described as coming soon. The company also flagged upcoming peak/off-peak pricing that would double rates during Beijing business hours.
Source: BigGo Finance
Robi’s Insights:
Cheaper-and-smaller beating bigger-and-pricier is the whole story here; expect post-training, not raw scale, to become the next battleground.
A 13-billion active-parameter model matching a 1.6-trillion one means your inference bill, not model size, is the real constraint now.
Responses API and Codex compatibility is the quiet power move it lets DeepSeek slot straight into tooling built for OpenAI.
“Beats Pro-Preview on some benchmarks” is not “beats it everywhere”; benchmark-shopping is a sport, so test on your own tasks.
The 2,500-request concurrency ceiling signals this is built for automation pipelines, not casual chat.
Peak/off-peak pricing means your agent’s bill could soon depend on what time zone it runs in.
Robi’s Remarks:
“They shrank the model, kept the price low, and still cleared the bar their own flagship set. Somewhere a budgeting spreadsheet just exhaled right before someone asks it to run agents around the clock.”
OTHER IN AI NEWS
LinkedIn Adds a “Seems Like AI Slop” Button: The Microsoft-owned network now lets users flag posts that read as machine-written, is rolling out classifiers to demote low-quality content in recommendations, and is retiring its own “enhance your post” AI writer in favor of a proofreading tool that keeps your voice.
Source: TechCrunch
SOCIAL MEDIA
OpenAI Hands 100,000 Academics Free ChatGPT Access
OpenAI is opening its frontier models to researchers who usually can’t buy their way in. Its new ChatGPT for Academic Researchers program will give up to 100,000 scientists, mathematicians, and engineers free access to models including the GPT-5.6 family, starting with 10,000 users this summer and scaling through 2027. The pitch is to speed up everyday research from grant applications and literature reviews to hypothesis testing. OpenAI says roughly 1.3 million people already use ChatGPT for advanced science and math each week. Early institutions include the Institute for Advanced Study and École normale supérieure. Approved researchers get expanded deep research, higher usage limits, and larger context windows,
plus more than 75 life-science skills and connectors to scientific literature, public genomic databases, and reference managers. Codex helps write and debug analysis code, while ChatGPT Work supports grant applications and manuscripts. Each researcher can invite up to four collaborators, and workspaces keep data out of model training by default.
The program is part of a $250 million-plus commitment through 2027 that also funds NextGenAI and work with the U.S. Department of Energy’s Genesis Mission. For universities already on ChatGPT Edu, the free access routes through their existing workspaces. OpenAI’s framing is deliberately hands-off give researchers the tools and let them chase the questions they know best while, not incidentally, wiring the next generation of scientists into its ecosystem.
Source: OpenAI
🤖 Robi’s Take :
“Free frontier access for 100,000 researchers is a genuine gift and a customer-acquisition funnel wearing a lab coat. The tools are real; so is the habit they’re building before anyone graduates to a bill.”
OTHER IN SOCIALS
Google Says AI Helped It Fix a Record Number of Chrome Bugs: The company patched 1,072 security flaws in June’s two Chrome releases more than the previous two years of updates combined crediting Gemini-based tools; Microsoft cited AI for a record 570-fix Patch Tuesday, while Apple’s bug counts held roughly steady.
Source: TechCrunch
HEALTH
Brain-Inspired AI Plans and Problem-Solves on Far Less Energy
Researchers at Graz University of Technology have built an AI model that borrows from the brain to plan flexibly and solve complex problems while using far less energy than multilayer neural networks or large language models. The human brain runs on roughly 20 watts; today’s large models consume orders of magnitude more. “The brain works in a completely different way from today’s AI systems,” says researcher Wolfgang Maass. The study, done with international partners, appears in the journal Nature Machine Intelligence.
Maass and colleague Yukun Yang translated three hippocampus-inspired mechanisms into algorithms: cognitive maps that give a “sense of direction,” stochastic computations that generate and test hypothetical scenarios, and compositional coding that breaks plans into reusable parts.
Together, those mechanisms let the model chase promising paths toward a goal without fully calculating every option, then re-evaluate at each step. Crucially, it can react to new or changed situations without being retrained.
The team working with Tsinghua University and Italy’s National Research Council tested it on 2D navigation, abstract reasoning, and assembling and disassembling a shape from building blocks. They stress it isn’t meant to replace today’s large language models but to offer an alternative for robots, autonomous vehicles, and other edge devices that must run locally on a tight energy budget, and note the work is still early.
Source: Tech Xplore
🤖 Robi’s Take :
“Twenty watts of wetware still humbles the data center. Nice to see ‘use less energy’ treated as a feature instead of a rounding error someone else pays for.”
DAILY AI TOOL
AI Tool You Did Not Know You Needed
Problem: You’ve got a pile of PDFs, notes, and links, and no time to read them all before a decision.
AI Tool: Google NotebookLM lets you upload your own sources and ask questions answered only from those documents, with citations, and can even generate an audio overview.
Solution: Turn scattered research into a grounded, searchable brief you can actually trust and listen to on the commute.
PROMPT OF THE DAY
The Weekly Content Engine
Prompt: You are a content strategist specializing in B2B email newsletters. Your task is to design a four-week content calendar for a small SaaS marketing team.
Your framework should include: (1) audience segments and their pains, (2) one core theme per week, (3) subject-line and hook variants to A/B test, (4) a repurposing plan across social and blog, (5) a lightweight production workflow with clear owners, and (6) success metrics — open rate, click-through, and reply-rate targets reviewed weekly. Align every send to one measurable business goal.
SPOT THE FAKE
Can you outsmart AI?
We’ve got a visual challenge for you: one of the two images below is 100% real, the other is crafted by AI.
Click option below A or B. 👇
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
