Glad you discovered the gem he is! What made me appreciate his expertise first? Most people use a dish washer wrong. His video will break it all down to help one understand how to use the typical dish washer in the right practical ways in which it was engineered.
I respect that the bling protest. I’d respect the writing to another level if they touch on some of the very public incidents that have happened in/around and associated with MIT, where this level of cost/security was even considered in the first place. This would open more dialog around better approaches as well.
Since I didn’t love the technical approach here (once one takes the humans out of the loop, there’s more ambitious things that can be done), I do appreciate the process. It’s not until the last sentence when process inspiration is revealed:
“Special thanks to Boris Cherny for encouraging us to be more ambitious.”
OpenAI is trying to level up their growth marketing to match Anthropic’s. Tactics in growth marketing (speaking from my experience leading in this space) are not public knowledge as it’s a cat-and-mouse game on multiple fronts (ie CAC goes up and changes the economics when a high ROI source is found, etc. etc. etc.), which makes hiring the right person vital as information asymmetries are where the ROI is (ie more revenue per dollar spent on marketing). Anthropic hired the right person for strategy and execution long ago in LLM-years, and there is a good podcast on the story here:
https://www.lennysnewsletter.com/p/anthropics-1b-to-19b-grow...
I think Dario is writing with much of these things in mind, but using this as a specific framed opportunity to enable a better interim and long-term outcome for this planet with us still on it. When the internal motives for Dario are fully shared on the other side of the hill, I think we will find his writing to be more about influencing and shaping policy to steer the world in the best way possible within his ability, control, and knowledge. It’s a commonly held believe with the effective altruism community, which Anthropic team was formed under, to use the policy levers to positively influence the world. If I read this latest writing from Dario through that lens, there are deeper concerns with the AI tooling that go well beyond what’s noted in this singular post and are likely the motivations for re-attempting a slowdown or pacing. I too generally agree with for many reasons like allowing the young OpenAI engineers to make mistakes like the Hugging Face incident in July and put better processes in place (ie where is your decent multi-level fencing controls and harness best practices when disabling security features and where is your energy/spend limits for achieving the mission objective in scope and not warming the world for no reason!) and more time to engineer and deploy better hardware to not warm the world instead of burning through fossil fuels since now data centers in the US can isolate pollution from the grid to exempt themselves from federal regulation.
That was sunset and replaced by Antigravity. FWIW until I abandoned it knowing the sunsetting, I was able to get good behavior out of Gemini CLI with overriding the system prompt. The default prompt crippled the harness with very poor instructions, but there was a hidden ENV to override it. Replacing it with Claude Code like prompts based on the model selected, it ran at a much higher intelligence level full stack with significantly less errors.
It may be a comforting idea to believe that our affection should require no assistance at all, and expect from all others that it should flow quite effortlessly from a caring mind. But human capacity, I have come to realize with both age, health, and genetic variations is rarely so simple a matter. When a person admits quite vulnerably that their memory falls short as the author has, yet their desire to care for their relationships remains entirely intact, there is something very moving in that alone. But to spend ten years quietly building as the author has is an act of love for people they care for are not lost to memory and forgotten in time.
“Meta internally projected late last year that it would earn about 10% of its overall annual revenue – or $16 billion – from running advertising for scams and banned goods, internal company documents show.”
Thanks for the clarification. I’m very happily wrong here, and I appreciate the correction. This does emphasize from an editorial review that the double meaning can be distracting for someone who isn’t very deep in the terminology of this particular area, so it may still be worth rewording the subheading for a broader audience which will be interested in this model.
A bit of editorial and cultural note from a US native, the subsection of the original article “Teach discrimination, not memorization” is better worded as something like 'Differentiation' or 'Distinction' instead of ‘Discrimination’. In English, the word 'discrimination' can (and in this social context may) imply social prejudice or unfair treatment. I think this may have been a bit of carry over from the rather benign French translation of “Enseigner la discrimination" which I also see awkwardly translated in the paper as well.
Having grown a large healthcare review platform, I can attest to the success we had mapping specific policy violations to natural language is incredibly useful. At scale, patients having terrible situations and/days can write about in ways that can be deeply unhealthy for the community or the doctors reading/receiving the feedback and sometimes very threatening beyond that purposes for the community. We built a custom ML engine to handle our levels of traffic for reviews, which was among the largest in the US typical ranking top 3 on Google for the domain keywords. Back when BERT was the edge, a policy-adaptive model like this one from Mistral would have been an incredible cold-start solution. Most sites never have the massive volume nor budget needed nor skillset needed before you can train domain-specific models that outperform OOTB solutions. Generally, most people and site mean well and try to do well, so empowering those people with models like this can help the collective in my opinion, so I’m happy to see this released in this manner myself.
This is specifically finance language for budgeting and one of the better ways to speak about this given the public disclosure of this information is substantial for their publicly traded stock.
I do exactly this (and more since my role is much broader and so is my approach) as a fractional head of product, data, and operations for multiple companies all in healthcare (fast growing self-funded to series D/IPOing soon). I saw your initial launch and felt validated by you all working on it, and now I’m further validated by the pivot. I have more work than I can handle, so I’m happy to share tips. You can find me via a bit of googling my HN handle or just adding a dot com to the end.
From inspecting the Cowork VM, the pollution is not documented and not controllable (publicly known - I have workarounds). It creates a lot of waste and frustration in the process.
CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD=1 means claude finds and loads all the CLAUDE.md of all the mounted repos overtime (and by settings). As such, working on multiple unrelated repos at the same time isn’t a pleasant experience out of the box.
A few other interesting VM ENVs:
CLAUDE_CODE_IS_COWORK=1
CLAUDE_CODE_BRIEF=1 CLAUDE_CODE_BRIEF_UPLOAD=1 CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 CLAUDE_CODE_DISABLE_BACKGROUND_TASKS=1 CLAUDE_CODE_DISABLE_CRON=1
CLAUDE_CODE_ENTRYPOINT=local-agent CLAUDE_CODE_EXECPATH=/usr/local/bin/claude CLAUDE_CODE_HOST_HTTP_PROXY_PORT=36543 CLAUDE_CODE_HOST_PLATFORM=darwin CLAUDE_CODE_HOST_SOCKS_PROXY_PORT=46673
USE_STAGING_OAUTH= _=/usr/bin/env all_proxy=socks5h://localhost:1080 ftp_proxy=socks5h://localhost:1080 grpc_proxy=socks5h://localhost:1080 http_proxy=http://localhost:3128 https_proxy=http://localhost:3128 no_proxy=localhost,127.0.0.1,::1,.local,.local,169.254.0.0/16,10.0.0.0/8,172.16.0.0/12,192.168.0.0/16
Can you push the results in a benchmark repo under the GitHub account linked from the parent Readme of the project?
The full transparency would be very helpful to know where these strengths are coming from which at a glance look to be multi-threaded in-memory processing.
Hi Fellow HN tech community, I’ve had the honor to digitally interact with Chad Whitacre, and from the first exchange to the subsequent ones over the months, he’s a good human. Chad provided a healthy, human-first approach to the most fundamental areas of free software. He possesses a deep understanding of how we all do better together. He is a person to celebrate for all the ways he was and is. Chad made our online lives richer by his ways of being. Cheers to Chad and his continued living as a good human!
Ivan worked at Align and was CTO of European DTC competitor. I agree with Ivan that both technologies have their sweet spot. I disagree with the scale piece and the frequency of visits as there are practices doing both those things differently using the placement and remote monitoring technology.
For younger patients that haven’t had braces before which is the majority of historical patient populations and future ones, 3D printed personalized braces systems are better as the tooth movement needed is more clinically substantial at tougher angles. Many younger patients forget and lose their plastic aligners.
That said: what is chosen should be what for the whole person at multiple levels.
This isn’t the case with 3D printed bracket systems. Studies have shown that the 3D personalized bracket system from LightForce will cut down treatment time over 40%. The refinement phase is also spot on and finishes much faster, and recent studies have shown further improvements. Also, the more complex the case, the larger the delta on the performance. Not all orthos are practicing this way but this is where the future is going whether it is LightForce or another company. The physics just works out better than the best plastics.
Disclaimer: I’m a health nerd and was an employee a couple years ago.
It's great to see that others have had better experiences than me. I had to upgrade from my M1 Air cause I kept on hitting issues. Note that I'm more on the power user side, and not on the typical light use side in my computer/software use cases day-to-day.
I would recommend a more systemic mapping like Power Scorecard from Demos.org [1] to help ground the mappings to impacts on populations. Ideally, one would be able to swap the system mappings based on the preferred context, as there are different policies that should be looked at in different ways to assure optimal positive impacts with minimal secondary negative impacts. I haven’t seen a perfect way of doing this yet, but this will be directional correct on the pathway to something even better for all of us. Thank you for your labor of love and efforts!
The founders are all on a first name basis. I’m surprised no one has noted that Anthropic and OpenAI winning together by giving the world two different choices, just like the US does in its political landscape. In this circumstance, OpenAI wins the local market for its government and aligned entities (while having the free consumer by a matter of cost dynamic for that ideal customer profile which is vary broad and similar to Google’s search audience where most their revenue still depends), while Anthropic is provided the global market and prosumer market where people can afford choice by paying for it.
My attempt with trying one of their OOTB prompts in the demo https://chat.inceptionlabs.ai resulted in:
"The server is currently overloaded. Please try again in a moment."
And a pop-up error of:
"The string did not match the expected pattern."
That happened three times, then the interface stopped working.
I was hoping to see how this stacked up against Taalas demo, which worked well and was so fast every time I've hit it this past week.
If we don't see a huge gain on the long-term horizon thinking reflected with the Vendor-Bench 2, I'm not going to switch away from CC. Until Google can beat
Anthropic on that front, Claude Code paired with the top long-horizon models will continue to pull away with full stack optimizations at every layer.
For a bit more context, most email infrastructures I’ve worked with are for transactional and marketing DTC and B2B companies. I would read my response in this context.
Re one-time setups and one-time changes: I think this will answer both questions and the implied PMF question as well. For internal FTE staff, this will be handle as a one off exception consistently (it’s really no one’s full-time job or responsibility). You may wish to speak with teams that offer professional services / SaaS including self-hosted where this infrastructure would be helpful. Their jobs are made easier with additional predicable / dependable infrastructure software (ie chat with (a) Twilio’s messaging team which remains the SendGrid acquisition, (b) related Red Hat / IBM) vs more work for an individual who is just doing this one-off. You may wish to consider a revenue share and/or white-labeling as they co-install the infrastructure for your business.