Skip to main content
Back

OpenAI launches Sora 2: physics-aware video with audio

Anthropic launches Claude 4.5 challenge, Thinking Machines Lab's research, DeepSeek's new model, Google AI Mode update.

October 1, 2025 4 min read

Hey

Welcome to AlphaSignal, the most read source of news by AI engineers and researchers.


Every day, we identify and summarize the top 1% of news, papers, models, and repos, so you’re always up to date.


Here’s today’s roundup:

Together with:

alpha_signal_image_2

Summary

Read time: 4 min 37 sec

Top News

▸ OpenAI launches Sora 2 with social app for video remixing

Delve

▸ Automate security questionnaires using Delve’s advanced agentic RAG platform

Signals

▸ Anthropic announces “Built with Claude 4.5” challenge for developers

▸ Ex-OpenAI CTO’s lab proves LoRA can equal full fine-tuning at lower costs

▸ DeepSeek drops open-source experimental model slashing API costs 50%

▸ Google enhances AI Mode, now supports uploaded images for visual search

▸ Replit introduces Connectors, letting apps natively use tools without credentials

Gladia

▸ Speed up speech recognition with Gladia’s sub-100ms partial transcripts

Trending Reads

▸ Anthropic explains why agents need careful context engineering strategies

▸ Microsoft: Can memory finally make AI agents self-improving?

▸ Simon Willison on when to build agentic loops; and when not to

How To

▸ Migrate to Claude Agent SDK from Claude Code SDK

Top News

OpenAI introduces Sora 2, producing physics-consistent videos with synchronized dialogue and sound

alpha_signal_image_3

OpenAI introduced Sora 2, its next-generation video generation model. The system is significantly more realistic than its predecessor, producing videos that obey physical laws. It can handle scene changes, and now include synchronized dialogue and sound effects.


Key Features

  • Physics-aware video: Models realistic outcomes, including misses, rebounds, and natural motion.

  • Synchronized audio: Generates speech, sound effects, and background audio aligned with video.

  • Controllability: Follows multi-shot instructions while maintaining consistent world state.

  • Cameos feature: Lets users insert their likeness and voice directly into generated scenes.

  • Styles supported: Handles cinematic, realistic, and anime video generation.

Performance

  • 5–10s outputs: Longer, coherent clips with complex scene changes.

  • High physical fidelity: Accurately simulates dynamics like buoyancy, rigidity, and momentum.

  • Challenging tasks: Handles Olympic-level routines, backflips on unstable surfaces, and detailed motion sequences.

  • Error realism: Missed actions behave naturally like ball bouncing off rim instead of teleporting.

Safety and Protections

  • Watermarking and provenance for all outputs.

  • Consent control: Users own and manage their cameo likeness.

  • Teen safeguards: Daily content limits, stricter cameo permissions, parental controls.

  • Moderation: Human and automated review for harmful or abusive content.

Deployment and Access

  • Sora app (iOS): Invite-only launch in the U.S. and Canada.

  • Web access: sora.com unlocks after app invite approval.

  • Android: Access via sora.com with invite code.

  • Pro tier: Sora 2 Pro model available for ChatGPT Pro subscribers.

  • API: Coming soon; Sora 1 Turbo remains supported.

Try now
alpha_signal_image_4

Turn RAG Into Riches With Delve’s AI for Security Questionnaires

Endless security questionnaires waste time. Delve’s compliance platform automates the grunt work using state-of-the-art agentic RAG, far beyond basic “retrieve + answer” approaches.


Key Features

  • Pull evidence, resolve conflicts, and reason across policy graphs

  • Interrogate infrastructure and draft bulletproof responses automatically

  • Trusted by Lovable, Bland, Micro1, and fast-growing AI companies

  • Proven track record navigating reviews with nearly every Fortune 50

  • Save dozens of hours on SOC 2, HIPAA, GDPR, and more

Book a demo with Delve and get $1,000 off SOC 2, HIPAA, GDPR and more - and automate security questionnaires away forever.

Book demo now

partner with us

Signals

Anthropic launches Claude Sonnet 4.5 contest, rewarding winners with one year Claude Max 20x and $1,000 API credits

1,846 Likes

Thinking Machines confirms LoRA achieves equivalent results to full fine-tuning at two-thirds cost

4,827 Likes

DeepSeek introduces V3.2-Exp, an open-source sparse attention system matching benchmarks at half cost

6,536 Likes

Google expands AI Mode in Search, letting users explore visually with conversational queries and uploads

2,7363 Likes

Replit presents Connectors, enabling one-click integration with services like Google, Salesforce, and Notion

621 Likes

Speed Up Real-Time Speech Recognition: Gladia Outputs Partials in <100ms

Start transcribing before you finish speaking. Gladia’s STT API streams partials with sub-100ms latency, improving voice agent responsiveness, intent understanding, and user experience.

Achieve up to 2x faster results than other providers offering partial transcripts.

Try it now ↗️

Reads

Anthropic explains why agents need careful context engineering strategies

2,423 Likes

The post explains context rot, attention budgets, and finite memory. It outlines practical techniques like compaction, structured note-taking, minimal tool design, and sub-agents to keep agents coherent and effective.

Microsoft: Can memory finally make AI agents self-improving?

743 Likes

Microsoft outlines how memory transforms stateless agents into adaptive ones. It explains working, short-term, long-term, persona, and episodic memory, with practical implementations using Mem0, Structured RAG, and Azure AI Search.

When to build agentic loops; and when not to

884 Likes

The post defines an LLM agent as “runs tools in a loop to achieve a goal.” Willison explores risks (YOLO mode), tool selection, sandboxing, and when loop design is appropriate.

How To

Migrate to Claude Agent SDK from Claude Code SDK

This guide shows how to update your codebase to the new Claude Agent SDK.


Step 1: Update Dependencies
Remove claude-code-sdk and install the new package:


npm uninstall claude-code-sdk
npm install @anthropic-ai/agent-sdk

Step 2: Update Imports
Replace old imports:


// Before
import  from "claude-code-sdk"

// After
import  from "@anthropic-ai/agent-sdk"

Step 3: Update Initialization

Change client setup:


// Before
const client =
    new ClaudeCode()

// After
const client =
    new ClaudeAgent()

Step 4: Update Method Calls

Refactor method names to match the new SDK:

  • client.runCode() → client.run()

  • client.reviewPR() → client.review()

  • client.chat() → client.converse()

Step 5: Test and Verify

Run your existing scripts or pipelines. Verify that:

  • PR reviews trigger as expected

  • Code execution completes without errors

  • Conversations work with the updated API

Read more

At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.


Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.

Looking to promote your company, product, service, or event to 250,000+ AI developers?

Work with us