Coregorithm

AI for everyone.

📩 [email protected]

[Coregorithm @coregorithm]

Volkswagen on track to cut 19,000 jobs by year-end.

Volkswagen is pressing ahead with one of the largest workforce reductions in its history, with CEO Oliver Blume set to confirm at the company's annual general meeting next Thursday that more than 28,000 positions will be eliminated across Germany by 2030.

[Coregorithm @coregorithm]

KPMG has pulled a major AI report after researchers discovered it contained fabricated case studies and hallucinated citations.

The report, titled "Total Experience: Redefining Excellence in the Age of Agentic AI," was published in October 2025 and cited major organizations including UBS, NHS Greater Manchester, Transport for London, and Swiss Federal Railways but the claims about all of them were either invented or wildly exaggerated beyond what the cited sources actually said.

AI detection group GPTZero found that out of 45 citations, only five were accurate. The rest were fabricated, paraphrased, or too vague to verify. Researchers believe KPMG staff used an AI research tool to generate the content without any human fact checking.

The incident is part of a growing pattern GPTZero had previously flagged similar hallucination issues in reports from Deloitte and Ernst & Young.

[Coregorithm @coregorithm]

Huawei showed off HarmonyOS 7 at its developer conference in China.

The visual update is called "Liquid Glass" translucent, light-reactive elements that feel similar to recent moves from Apple and Samsung.

What actually matters more is the new agentic AI system. It’s designed to take on multi-step tasks across different apps on its own. Huawei says it links up with over 2,000 specialized agents and gets above a 90% success rate on complex jobs.

The XiaoYi assistant also got an upgrade; it now holds onto context instead of resetting with every new question.

Performance is up around 15% from the last version, and the OS stretches across phones, tablets, PCs, wearables, and smart home devices. The release comes at a convenient moment Apple has been slow to roll out meaningful AI features in China because of regulatory friction. A wider public version should arrive this fall.

[Coregorithm @coregorithm]

Kimi K2.7-Codevis Moonshot AI’s new open-source coding model. It’s not just another model, it’s built to be smart without overthinking.

Why it’s beneficial:
- It uses 30% fewer reasoning tokens. This means it gives you answers faster and cheaper.
- Strong performance in real coding tasks, multi-language work, and tool use.
- Very affordable API ($0.95 per million input tokens).
- Fully open-source and works with the same tools you already use (OpenAI & Anthropic SDKs).

What’s it actually useful for?
- Writing, debugging, and improving code
- Building AI agents and tools
- Long coding sessions where normal models waste too many tokens thinking
- Developers who want good results without high costs

Why you should use it:
In 2026, speed and cost matter more than raw size. Most models overthink and burn tokens (and your money). Kimi K2.7 is trained to **stop overthinking** — it delivers cleaner, faster results especially in coding and agent work.

If you code, build tools, or run agents, Kimi K2.7 gives you better efficiency and lower cost than most current options. It’s one of the smartest “practical” coding models right now.

[Kilo @kilocode]

https://t.co/XY4uaKmYFl

[Coregorithm @coregorithm]

Jeff Bezos has launched a new AI company called Prometheus.

While most AI tools today focus on writing text or creating images, this one is built to help design and build real physical things like machines, engines, buildings, or complex products.

Right now, creating something complicated (for example, a jet engine or a skyscraper) usually takes a big team of engineers and many years. Prometheus wants to make this process much faster and cheaper. Their goal is to reduce the time and number of people needed dramatically potentially from 100 engineers working for 10 years down to just 10 engineers finishing it in about a year.

The company is working on an AI system that actually understands physics, materials, and how things are manufactured, not just words. It’s designed to work like a very advanced engineering assistant that can help with the entire process of designing and building physical products.

Prometheus was started by Jeff Bezos together with Vik Bajaj (who previously worked at Google X). The company already has around 150 employees and offices in San Francisco, London, and Zurich. It has raised a huge amount of money $18.2 billion and is currently valued at $41 billion.

In short, if tools like ChatGPT changed how we create writing and content, Prometheus wants to do something similar for designing and building real-world objects.

#ai #prometheus #jeffbezos

[Coregorithm @coregorithm]

A Munich court has ruled that Google can be held directly liable for false information generated by its AI Overviews feature.

The case was brought by two local publishers who were wrongly linked to scams and questionable business practices in Google’s AI-generated summaries.

The court found that these AI summaries count as Google’s own statements, so the usual legal protections that apply to regular search results don’t cover them. Google was ordered to stop showing the inaccurate claims and must pay 80% of the legal costs. The company has said it disagrees with the decision and plans to appeal.

The ruling could influence how courts elsewhere treat liability for AI-generated content in search.

[Coregorithm @coregorithm]

Meta is going all in on AI, and the shift is causing real turbulence inside the company.

They've laid off around 8,000 people and moved another 7,000 into AI-focused roles. A lot of employees are anxious about what this means for their jobs and the company's direction.

In an internal message, Mark Zuckerberg acknowledged that Meta had made mistakes and apologized. He also tried to reassure people that there won't be any more major layoffs this year.

The hard part is the money. Building serious AI infrastructure at this scale is incredibly expensive some estimates put the total cost as high as $145 billion and it's still unclear whether the investment will actually pay off.

[Coregorithm @coregorithm]

More developers are quietly moving toward local AI models. Running models on your own computer or server gives more control and stability. After recent changes with big closed models, this option suddenly looks more attractive. Local AI is no longer just for hobbyists. It is becoming a serious choice for people who want to keep building without depending on one company.

[Coregorithm @coregorithm]

Google DeepMind just released DiffusionGemma (June 10), and it actually feels a bit different from the usual updates.

It’s built on the Gemma 4 26B MoE backbone, but instead of predicting text one token at a time, it uses a diffusion approach. It starts from noise and gradually refines whole chunks of text at once, kind of like how image diffusion models work.

The interesting part is what that unlocks. It can be up to four times faster on GPUs, which suddenly makes real-time local use much more practical. Running something like this on a high-end laptop doesn’t feel out of reach anymore, especially if you pair it with tools like Unsloth to squeeze out even more speed.

There is a trade-off, though. You give up a bit of factual precision, so it’s not ideal for heavy reasoning or tasks where accuracy really matters. But for creative work, prototyping, or just fast interactive use, it actually fits really well.

It feels like an early glimpse of a shift. Instead of only chasing bigger models, we might start seeing more focus on new ways of generating outputs that are optimized for speed, responsiveness, and running locally.

Definitely curious to see what the open-source community builds with this.

#google #gemma

[Coregorithm @coregorithm]

Fable 5 is now offline. Anthropic says the US government ordered access to Fable 5 and Mythos 5 to be suspended for foreign nationals, and Anthropic then disabled both models for all customers to comply. Anthropic says it wasn’t given the exact security reason, but believes it may relate to a reported “jailbreak” method.

[Coregorithm @coregorithm]

This acquisition reflects a shift in focus from model performance alone toward the full stack required for reliable agents especially around long-horizon reliability, state management, and secure deployment. It’s less about making models smarter and more about making existing intelligence usable at scale in real environments.

[Coregorithm @coregorithm]

OpenAI’s acquisition of Ona is easy to overlook, but it targets a real bottleneck in agent development: persistent and secure execution.

Most agent demos still run in short browser sessions. For agents to handle longer, real-world work, they need to maintain state over hours or days, keep running when you’re offline, and meet enterprise security and compliance standards.

By integrating Ona into Codex, OpenAI is strengthening the layer that allows agents to move beyond helpful assistants toward systems that can reliably manage extended tasks. Model intelligence is advancing quickly, but without a solid execution environment, even capable models remain limited to demos.

Infrastructure moves like this rarely make headlines, yet they often determine how far agentic systems can actually go in production.

[Amazon Web Services @awscloud]

More AI-generated code doesn't make your team faster. It might actually slow you down.

[Coregorithm @coregorithm]

@testerlabor That's true, however, Grok disappointed me.

[Coregorithm @coregorithm]

Did you enable any memory related settings? I had same issue before check this out;

On https://t.co/7W1uLmzStk or the Grok app, click on your profile picture (top right).
• Look for Settings → Memory or Personalization.
• There you can usually see, edit, or delete saved memories.

Enable if it's closed that section

[Coregorithm @coregorithm]

@npmlol Source ?

[Coregorithm @coregorithm]

Even though current AI models are still far from true human-level intelligence, they’re already becoming extremely expensive to run at scale.

Fable 5 is a good example. It’s genuinely powerful on complex tasks, but it’s also very token heavy and costly compared to previous model

This creates an interesting tension:
If AI keeps getting more capable but also more expensive to operate, how is it actually going to replace human skills and labor at scale?

Replacing humans economically only makes sense if the AI is cheaper (or at least comparable) in the long run not significantly more expensive.

I’m curious how people see this playing out.
Do you think:

Pricing will drop dramatically with scale and optimization?

AI will mostly augment rather than replace humans because of the cost?

Or will it only replace high value work where the output is worth the higher cost?

Would love to hear different perspectives.

#ai #claude #job

[ClaudeDevs @ClaudeDevs]

Claude Fable 5 is here. New model generation, new way of working.

Here's how to get started in Claude Code and on the Claude Platform: 🧵