Yeah, so it’d be interesting to see if provided the correct context/your understanding of its error pattern, it can accomplish this.
One thing you learn quickly about working with LLMs if they have these kind of baked-in biases, some of which are very fixed and tied to their very limited ability to engage in novel reasoning (cc François Chollet), while others are far more loosely held/correctable. If it sticks with the errant patten, even when provided the proper context, it probably isn’t something an off-the-shelf model can handle.
I also experienced recently that gate agents will lie about this to decrease turn around time if a plane is running late. We were forced to gate check ours after being told the plane was too full to hold all, only to get onboard and find ample space. When we asked the flight attendant, they apologized and said they did not tell them to do this, but it’s unfortunately common at airports with frequent delays. So charge ridiculous bag fees, forcing everyone to carry on, which delays boarding, so they lie to you and check your bag for free (which maybe you would’ve done to begin with if that was an option), only to board a plane and feel duped. All around great experience!
Yes, because the UK’s two dominant (and right wing) parties have been actively sabotaging it for years, chasing after a despicable dream of homegrown middlemen and fraudsters, envious as they are of the unchecked criminality of their friends from across the pond. Quelle surprise, things have gotten worse.
> The US spends more per capita on their social safety net than almost all other countries, including France and the UK.
To a vast and corrupt array of rentiers, middlemen, and out-and-out fraudsters, instead of direct provision of services, resulting in worse outcomes at higher costs!
Turns out if I’m forced to have visits with three different wallet inspectors on the way to seeing a doctor, I’ve somehow spent more money and end up less healthy than my neighbors who did not. Curious…
Are you saying Claude is better at writing Rust or Gemini? Or Gemini in Gemini CLI? Not following. In my experience, Gemini 2.5 PRO is better at writing Rust overall.
Gemini 2.5 Pro is a better coding model, but Gemini CLI is way behind Claude Code, perhaps because the model itself isn’t well-tuned for agentic work. If you’re just doing targeted refactoring and exploration, you can copy and paste back and forth from the web app for $20 a month.
They could, they’re just greedy, self-serving, and short-sighted. China’s doing the AI equivalent of Belt and Road to reap tremendous strategic advantages, as well as encourage large-scale domestic innovation.
I’m not sure this follows as SOTA LLMs are pretty good at writing Rust, so wouldn’t this also make it easier for codebases to move leftward in your analogy? For example, I was resistant to use Rust for a lot of things because a). The community is somewhat annoying and pedantic, even by software engineering standards b). The overhead in getting colleagues up to speed on Rust code was too much of a time suck. LLMs solve both those problems and we’re now migrating lots of stuff to Rust, my colleagues can ask lots of questions (of Google Gemini Pro 2.5), without burdening anyone or being met with disdain, and seem generally more curious and positive about these moves/Rust overall.
I’m blown away by how good Gemini Pro 2.5 is with Rust. Claude I’ve found somewhat disappointing, although it can do focused edits okay. Haven’t tried any of the o-series models.
Yeah, my experience of working with Claude Code is that I’m actually far more conscientious about design. After using it for awhile, you get a good sense of its limits and how you need to break things down and spell thing out to overcome.
Neoliberalism and globalization were the state ideology of the American empire, of which the EU (was) a part, so of course they’re experiencing the same. They’ll now either re-align themselves with China (who capitalized best on the prior era) or face a similar fate.
We’re watching the fire sale of America, like was imposed on Russia in the 90s, and resulted in one of the largest declines in life expectancy in the country’s history. I expect the same to happen here, including its eventual culmination in the rise of a Putin-like figure from the security state apparatus, after we similarly suffer a decade or more of internal collapse and humiliation.
> which is lower valued, and thus it is economically "correct" to have them be replaced when an appropriate automation method is found.
Textbook example of why this “economic” form of analysis is naive, stupid, and short-sighted, (as is almost always the case).
AI models will never obtain the ability to completely replace “low value work” (as that is not perfectly definable or able to be defined in advance for all cases), so in a scenario where all engineers devoted to these tasks are let go, what you would end up with is a engineers higher up the value chain being tasked with resolving the problems that result from when the AI fails, underperforms, or the assessment of a task’s value was incorrect. The cumulative effect of this would be a massive drain on the effectiveness of said engineers, as they’re now tasked with context switching from creative, high-value work to troubleshooting opaque, AI code slop.
I haven’t tried with the most recent Claude models, but for the last iteration, Gemini was far better at Rust and what I still use to write anything in it. As an experiment, I even fed it a whole ebook on Rust design patterns and a small script (500 lines) and it was able to refactor to use the correct ones, with some minor back and forth to fix build errors!
Gemini 2.5 Pro is better at coding than Claude, it’s just not as good at acting agentically, nor does Google have good tooling to support this use case. Given how quickly they’ve come from far behind and their advantage on context size (Claude’s biggest weakness), this could change just as fast, although I’m skeptical they can deliver a good end user dev tool.
This is generally controllable with prompting. I usually include something like, “be excessively cautious and conservative in refactoring, only implementing the desired changes” to avoid.
It’s annoying, but I’ve done extensive work with this model and leaving the comments in for the first few iterations produced better outcomes. I expect this is baked into the RL they’re doing, but because of the context size, it’s not really an issue. You can just ask it to strip out in the final pass.
Claude Code works fairly well, but Anthropic has lost the plot on the state of market competition. OpenAI tried to buy Cursor and now Windsurf because they know they need to win market share, Gemini 2.5 pro is better at coding than their Sonnet models, has huge context and runs on their TPU stack, but somehow Anthropic is expecting people to pay $200 in API costs per functional PR costs to vibe code. Ok.
What’s the best library for fine-tuning VLMs at the moment and do they support this architecture or that for the IBM Granite vision models? Document understanding tasks seem in special need of fine-tuning.
What I always find hilarious about these naive libertarian types is they never even bother to check their hypotheticals against reality. For example, FutureMotion had to have a regulatory body intervene because they were killing and injuring people with their skateboard designs: