F U Z Z Y G O A T ™

Digital nomad, computer scientist, and AI specialist with a background in computer graphics and visual effects, developing @NeoTorino and @NeoGenBase

[F U Z Z Y G O A T ™ @Fuzzygoat]

The greatest competitive advantage of the next decade won’t be using AI. It’ll be having access to the AI that nobody else does.

[F U Z Z Y G O A T ™ @Fuzzygoat]

@ListerLawrence Always walk the water, it’s 101 of driving off-road.

[F U Z Z Y G O A T ™ @Fuzzygoat]

Sorry to hear that. I know how you feel.

Back in the early 2000s, my employer sent me to SIGGRAPH because they wanted to improve the studio’s 2D and 3D pipeline. A month after I got back, they shut the studio down.

To make matters worse, they argued over redundancy, overtime, and holiday pay, then deducted the entire cost of the SIGGRAPH trip from my final paycheck.

They were a good company, which almost made it more surprising. It taught me a lesson I’ve never forgotten: no matter how decent a company seems, never underestimate how low it might go when money is involved.

All the best in finding your next position, and good luck on your next adventure. ❤️

[F U Z Z Y G O A T ™ @Fuzzygoat]

I think the recent restrictions on frontier models by @OpenAI and @AnthropicAI mark a watershed moment that will reverberate across the AI industry for years to come.

People are beginning to realize just how unstable and unpredictable the frontier model landscape has become. A combination of corporate hubris and increasing government influence, often justified on national security grounds, is steadily eroding the commercial value of these systems. After all, there’s little point in having the world’s best model if you can’t reliably put it into the hands of your customers. Restricting access undermines trust, weakens adoption, and ultimately hampers long-term growth.

By contrast, open-source models may not always match the absolute frontier in raw capability, but they offer something businesses value just as highly: stability, consistency, and control. You can’t build products or companies around a model that may be available today and withdrawn tomorrow.

The frontier labs also face another challenge. The smartest people aren’t always the ones on the payroll. Many of the most creative breakthroughs come from independent researchers and hobbyists who build simply because they enjoy solving hard problems, not because they’re chasing stock options or headlines. Those quiet innovators are the lifeblood of open source.

That’s why I believe open-source AI will continue to thrive. It may not always be the most powerful, but for many real-world applications it will become the platform people trust when they simply need reliable intelligence that gets the job done. #ai

[F U Z Z Y G O A T ™ @Fuzzygoat]

Víkurfjara

[F U Z Z Y G O A T ™ @Fuzzygoat]

The Sol/Mythos release may have changed expectations permanently. Once people realise that the public model isn’t necessarily the most capable model, the incentive shifts.

Instead of releasing true frontier systems, AI labs may increasingly present an acceptable public version while keeping the real breakthroughs behind closed doors for governments, major enterprises, and a handful of strategic partners.

That creates a two-speed AI world. The public gets a reasonable model, but not the best one. Meanwhile, those with privileged access gain an enormous advantage in research, engineering, healthcare, finance, defence, and virtually every other knowledge-based field.

The biggest competitive edge in the next decade may not come from who uses AI, but from who has access to the AI that nobody else can use. @OpenAI @AnthropicAI

[F U Z Z Y G O A T ™ @Fuzzygoat]

The great bait and switch has started …. After this release (Sol / Mythos) where the public now know they’re being served watered-down models, I expect frontier AI labs to pivot toward a smoke-and-mirrors strategy.

The “latest and greatest” will be marketed to everyone, but behind the scenes it won’t actually be the most capable model. Instead, the real frontier systems will be reserved for a small group with privileged access, major corporations, governments, and strategic partners.

The result is a two-tier AI ecosystem: one model for the masses, another for the insiders. Those with access to the genuine frontier models won’t just have a better chatbot; they’ll have a significant competitive advantage in science, engineering, business, and defence, while everyone else is left believing they’re using the cutting edge. @OpenAI @AnthropicAI

[F U Z Z Y G O A T ™ @Fuzzygoat]

@BlueWestlo Wait for the PvE mode.

[F U Z Z Y G O A T ™ @Fuzzygoat]

@ddrkeby Very very nice ❤️

[F U Z Z Y G O A T ™ @Fuzzygoat]

@wardonkey Love the animations, Sentinel has a great feeling of weight a presence, exceptional work.

[F U Z Z Y G O A T ™ @Fuzzygoat]

Arc Raiders is doing just fine. It’s only been out for nine months, we already have a solid selection of maps, the new modes and features from the Chinese beta look genuinely promising, and we know a major update is coming in October, just four months (16 weeks) away. On top of that, new map conditions also appear to be on the horizon as part of the ongoing live service.

The expectations some people have for live-service games are completely unrealistic. They play games as a full-time hobby and expect developers to endlessly feed their content pipeline, while being the first to tear apart anything that isn’t 100% perfect.

Game development takes time. A single artist can easily spend a couple of weeks creating and texturing a character skin. A new Arc can take months to design, prototype, balance, and polish. An entirely new map can take a full team six to nine months from concept to release.

So just relax and let them cook. If you’ve run out of things to do for a while, that’s fine, play something else and come back when the next update lands. It’s not @EmbarkStudios job to fuel your content pipeline, that’s your job.

[OpenAI @OpenAI]

Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.

https://t.co/OoM83SyISN

[F U Z Z Y G O A T ™ @Fuzzygoat]

I don’t think the AI race is going to be good for the world. Too many states have been caught with their collective pants down, leaving them years behind in what may become the defining technology of the century. But you have to respect the position the USA is putting itself in, it’s going to turn out to be an extraordinary strategic advantage.

[F U Z Z Y G O A T ™ @Fuzzygoat]

@awkwardgoogle Just before getting hit by a HGV driver watching YouTube on his phone. 🤷‍♂️

[F U Z Z Y G O A T ™ @Fuzzygoat]

Frontier-model distillation could become the AI equivalent of an underground black market. Engineers will be stuffing encrypted drives into duffel bags, wiping servers, and disappearing into the night as special forces descend to seize GPU clusters. The world’s most valuable contraband won’t be gold or drugs, but compressed intelligence, model weights that simply fit in a backpack.

[F U Z Z Y G O A T ™ @Fuzzygoat]

@GavinNewsom Yet people who are in the employment of billionaires are making a good living.

[F U Z Z Y G O A T ™ @Fuzzygoat]

@haider1 I think that may well have been the plan all along, to derail the whole thing. He just happened to stumble into the most capable model, which complicated things.

[F U Z Z Y G O A T ™ @Fuzzygoat]

@JamesMelville Someone, the person or org behind the charts is pushing a narrative, rather than presenting useful data.

[F U Z Z Y G O A T ™ @Fuzzygoat]

Primarily because the only colour they seem to use above 30°C is red. 😂 According to the BBC, there are effectively only two temperatures on the map. A more balanced, less exaggerated colour gradient would show far more useful variation, helping people actually plan their day and stay cool, rather than making half the map look identical.

[F U Z Z Y G O A T ™ @Fuzzygoat]

@NorgannonHH No, with Ermals slots you can pack any blueprint you like, with the expidition to get 5 random blueprints (all unique). That may or may not duplicate one that you have already packed.