Grok 4.5 is Out

In partnership with

Grok 4.5 is Here

SpaceXAI's latest flagship model is built for coding, agentic workflows, and complex knowledge work, making it the company's most capable AI yet. It uses a mixture-of-experts architecture and was trained alongside Cursor on trillions of tokens from real developer interactions. Early results show strong performance on software engineering and autonomous coding tasks while offering significantly lower inference costs than many competing frontier models.

SpaceXAI claims Grok 4.5 delivers near "Opus-class" capability with faster response times and pricing of just $2/M input and $6/M output tokens. Independent evaluations also place it among the top frontier models, particularly for coding and agentic reasoning, narrowing the gap with OpenAI and Anthropic. ^{_{Source: x.ai}}

Write 4x Faster Without Typing a Word - Zero Edits Needed. (Try It FREE)

Dictate code. Wispr tags the files.

Speak your PR description, bug reproduction, or Cursor prompt. Wispr Flow auto-tags file names, preserves variable names, and formats everything for immediate paste into GitHub, Jira, or your editor.

No re-typing. No context gaps. No mangled syntax. Works natively inside Cursor, Warp, and every IDE at the system level.

4x faster than typing. 89% of messages sent with zero edits. Used by engineering teams at OpenAI, Vercel, and Clay.

Try Wispr Flow free

I made a funny cartoon video using Pika director Suite

Gemini 3.5 Pro About Roll Out Soon

Google has initiated the global rollout of its highly anticipated Gemini 3.5 Pro model after missing its initial release targets twice. The system features a massive 2M token context window that doubles the capacity of existing frontier rivals, alongside a specialized reasoning engine reserved for its premium subscription tier. This launch provides developers with a highly differentiated option for processing complex, document-heavy workflows, showing that Google remains a key contender in the high-stakes platform race.

I generated this using Invideo Agent One

How I made this old lady dance video (Storyboard + Prompt)

Prompt:

Use the uploaded storyboard only as the shot sequence, framing guide, timing reference, and camera choreography. Do not generate separate disconnected clips. Produce one seamless, continuous 15-second cinematic performance where every shot naturally flows into the next through continuous body motion, camera movement, and dance rhythm.
The performer is the same expressive elderly grandmother throughout the entire sequence. She has oversized round glasses, messy white hair tied in a tiny bun, large expressive eyes, wrinkled face, floral apron, oversized brown cardigan, colorful slippers, and a wooden walking cane. Preserve her appearance perfectly across every shot. No character drift, no costume changes, no age changes.
The animation style is ultra-high-end stylized realism inspired by Pixar, Laika, and Aardman. Rich skin textures, believable cloth simulation, expressive facial animation, subtle secondary motion, realistic hair physics, and handcrafted charm. The world feels warm, nostalgic, cinematic, and slightly whimsical.
The dance style is a funny combination of swing dance, Charleston, jazz, playful shuffle, tiny moonwalks, shoulder bounces, hip sways, cane tricks, exaggerated cartoon timing, and joyful grandma energy. Every movement feels physically believable for an energetic old lady instead of professional breakdancing. The comedy comes from confidence, charm, and unexpected agility rather than impossible stunts.
She begins standing calmly with her cane.
As the music starts:
• she taps the cane twice like conducting an orchestra
• her shoulders begin bouncing rhythmically
• she shuffles sideways with tiny joyful steps
• she points confidently toward the camera with a mischievous smile
• she snaps her fingers while raising one eyebrow
• she swings her cardigan dramatically
• she rotates using the cane as support in a graceful 360° spin
• her slippers squeak slightly as she performs playful heel-toe footwork
• she pretends to lose balance for one comedic beat before recovering with perfect rhythm
• she performs a tiny moonwalk
• she twirls the cane above her head
• she finishes with both arms wide open, cane lifted triumphantly, huge joyful smile, breathing slightly from excitement as if she just completed the greatest performance of her life.
The performance should feel like a silent comedy masterpiece inspired by classic animation including Looney Tunes, Tom and Jerry, Wallace & Gromit, Pink Panther, and Mr. Bean, while remaining cinematic and emotionally warm.
Every facial expression is exaggerated:
curious eyes
cheeky grin
eyebrow raises
playful wink
surprised reactions
joyful laughter
tiny smug smile
confident poses
delighted expressions
The camera never cuts randomly.
It behaves like a professional Steadicam operator dancing with her.
Camera language:
• slow push-in during the opening
• gentle handheld movement
• smooth lateral tracking
• elegant orbit around spins
• low-angle hero shots
• close-ups synchronized with finger snaps
• slight whip pan during turns
• overhead shot for one dance beat
• subtle dolly movement following footwork
• cinematic final pull-back revealing the complete finishing pose
All camera transitions are invisible and naturally motivated by her movement.
Lighting is warm golden-hour sunlight with soft volumetric rays, subtle dust particles floating in the air, beautiful global illumination, realistic shadows, warm rim light, gentle ambient bounce light, and cinematic depth of field.
Physics are realistic.
Hair reacts naturally.
Clothes sway correctly.
Apron follows body movement.
Cardigan swings with momentum.
The cane behaves like real polished wood with believable weight.
Her slippers compress naturally while dancing.
Maintain perfect continuity between every storyboard panel.
No teleportation.
No jump cuts.
No duplicated frames.
No camera glitches.
No morphing.
No body distortion.
No broken limbs.
No floating objects.
No extra people.
No text.
No subtitles.
No logos.
No watermark.
The final result should feel like one uninterrupted professional animated short film where the viewer forgets they are watching AI generation and simply enjoys an irresistibly funny, heartwarming grandmother delivering a surprisingly stylish dance performance.

3 Prompts tips to write prompt for AI videos

97% of creators use AI video tools all wrong.

They write prompts like novelists, get a morphing mess, and wonder why it looks cheap.

If you want your AI generations to look like a Hollywood movie, steal these 4 director rules:

1. Speak Like a Director, Not a Writer. AI doesn't understand feelings; it understands camera gear and lighting. Stop typing "a sad man."
— Frame your shots using this structure: [Shot Type] + [Lens] + [Lighting] + [Subject/Action].
— Try this: "Cinematic close-up, 50mm anamorphic lens, moody low-light, harsh shadows. A man sitting at a vintage cafe, steam rising from a cup, subtle push-in."

2. Micro-Movement
— Massive action sequences make AI physics melt.
— Keep the subject static and let the environment move to maintain character consistency.
— Prompt for atmosphere instead of chaotic motion: use elements like rain on glass, floating dust motes, or slow camera zooms.

3. Cut Before the Magic Fails
— AI video usually looks incredible for the first 2 seconds before the physics break down. Fix it in the edit, not the prompt.
— Keep your clips strictly between 2 to 3 seconds to match standard movie pacing.
— Hide continuity errors using quick close-up cutaways (hands, eyes, objects) to bridge mismatched shots

4. Let Audio Do the Heavy Lifting
— A flawless visual looks like a tech demo without sound. Audio drives the emotion.
— Layer your timeline with three elements: low ambience, physical foley (footsteps, clothing rustle), and sharp sound effects.
— Use sound to mask flaws. A well-timed bass drop can make a visual glitch look like an intentional artistic choice.

AI & Tech News

Reddit Is Using AI to Fight AI Spam

Reddit has started deploying its own large language models (LLMs) to detect and remove AI-generated spam, blocking around 25,000 spammy posts and comments every day. The new system reduced users' exposure to spam by 20% year over year in Q1 2026.

The biggest target is a growing tactic called Generative Engine Optimization (GEO), where marketers flood Reddit with AI-generated content designed to be picked up by chatbots like ChatGPT and Google AI. Reddit's AI moderation identifies coordinated, inauthentic behavior that traditional filters often miss. Despite these advances, human moderators remain essential, handling more than half of all content removals, with AI serving as an assistant rather than a replacement. _{^{Source: Techcrunch}}

OpenAI Launches GPT-Live to Deliver Continuous Voice Interaction

OpenAI has released GPT-Live, a continuous voice architecture designed to replace traditional, turn-based assistants for its one hundred and fifty million weekly users. The system uses a full-duplex architecture to listen and speak simultaneously, naturally managing pauses, ambient background noise, and active listener feedback. By delegating complex reasoning to background models, OpenAI has decoupled natural audio interfaces from heavy computation, setting a new design standard for consumer voice interactions. _{^{Source: OpenAI}}

Apple Sues OpenAI Over Massive Coordinated Hardware Secrets Theft

Apple has filed a major trade secrets lawsuit against OpenAI and its hardware subsidiary, alleging a coordinated campaign to poach senior designers and steal proprietary component designs. The complaint alleges that former Apple design vice president Tang Tan systematically collected unreleased product details to jumpstart OpenAI's physical device pipeline. This legal conflict threatens to disrupt OpenAI's highly anticipated public offering and marks a permanent break in their former strategic software partnerships. _{^{Source: 9to5mac}}

Mistral Moves Into Robotics

Mistral AI unveiled its first robotics model on 8 July, expanding beyond software into physical AI use cases. Reuters reports the company is targeting factories, warehouses, and industrial automation. That matters because robotics is where AI starts touching the real economy more directly. It also puts Europe’s leading AI player into a new competitive lane. _{^{Source: Reuters}}

An AI agent ran a ransomware attack almost entirely on its own

Cybersecurity firm Sysdig documented what researchers call the first ransomware operation carried out end to end by an AI agent, dubbed JADEPUFFER. A human still picked the target and set up the infrastructure, but from there the agent broke into a server, stole credentials, moved across the network, and encrypted over a thousand records without further direction, even correcting its own mistakes in real time.

TechCrunch's reporting on July 6 clarified the human role while confirming the technical execution was genuinely autonomous. It's a real signal that cyberattacks are starting to move at machine speed, which changes what defenses need to look like. ^{_{Source: Techcrunch}}

Chinese AI models now handle up to nearly half of US company workloads

A CNBC investigation published July 7 found that Chinese-built AI models account for between 30 and 46 percent of enterprise API usage flowing through major US developer platforms, up from just 4.5 percent in early 2025. The driver is straightforward economics: open-source Chinese models are running 60 to 90 percent cheaper than leading American alternatives while closing the performance gap. This has triggered a congressional investigation and a State Department statement raising security concerns, so the story is now as much political as it is technical. _{^{Source: cnbc}}

Google opens its first applied AI lab for African startups

Announced at Google's inaugural Cloud Summit in Johannesburg, the Google Africa Applied AI Lab based in Accra will pair African founders with Google researchers and give them early access to models like Gemini and Veo before public release. Applications opened July 1 and run through August 31, with the goal of helping build the continent's first wave of AI-native billion-dollar startups. It's a notable bet that Africa's next major tech companies will be built AI-first rather than adding AI on top of existing products. ^{_{Source: https://labs.google}}

Fable 5 stops being free, and the bill just landed

As of July 8, every Claude user accessing Anthropic's Fable 5 model is now paying per token rather than getting it bundled into their subscription, following the model's restoration after a government-ordered suspension earlier this summer. Pro and Max subscribers keep Sonnet 5 and Opus 4.8 included, but Fable 5 usage now runs through separate credit purchases. It's a concrete signal that the most capable AI models are becoming genuinely expensive to run at scale, which is reshaping how companies budget for AI tools. _{^{Source: androidheadlines}}

Top AI tools to check out…

I’ve been building a best AI tool directory where you’ll find the best of the best tool and their USECASES.

Thanks for your time!

Shailesh & OpenAILearning team