Hustling for tomorrow’s tech!
Philadelphia
Joined May 2023
- Tweets1.3K
- Following269
- Followers20
- Likes2K
Alex Gardner retweeted
I love being able to switch models on @NousResearch.
It's a small thing but I love it. Well maybe its a big thing.
Can't do it with other AI agents that are LOCKED into a single lab
Alex Gardner retweeted
Sonarr. Radarr. Lidarr. Prowlarr. Bazarr. Seerr.
Six apps to configure, connect and maintain.
HolyArrs aims to bring that workflow into one Netflix-style interface for your own media library.
Browse, request, find releases and organize files. Connect your sources and downloader, then use Plex/Jellyfin/Emby for playback.
Go + Svelte + SQLite.
~15 MB compressed Docker image, according to the developer.
Free, open source, self-hosted.
github.com/HolyArrs/HolyArrs
Alex Gardner retweeted
overheard from friends talking about AI in China, and it’s very alarming
- AI short films went massive in China market because of how good the video models are now
- the first wave of people who mastered using AI to generate short films and comics made quite some money because they could produce a lot more content than traditional workflows
- but within just a few months, everybody and their mom could make equally good short films because all the expertise could be written in a markdown file
- when everyone’s more productive, no one actually has an advantage and it’s a race to the bottom. very quickly it became impossible to make money, and everyone’s busier than they were without AI
- the only one making money now is companies like bytedance who owns user demand and distribution. they sell ads to creators who pay to get traffic, and eventually ad spend became 80% of those creators’ operating expense, while AI tokens is less than 5%
so what are the problems here?
1. there’s no moat if you are creating anything with AI
if a human is really good at something, and it’s hard to find more humans who can do the same, they can leverage the scarcity of their skill to get paid very well
but if AI is good at something, it will be infinitely replicated, so there’s no scarcity hence you can’t really expect a sustainable business by creating anything with AI
2. more creators created more content people actually liked. that’s a good thing
but the economical benefit of that all got absorbed by the distribution platforms
there’s a famous chinese poem 遍身罗绮者，不是养蚕人 - it basically says people who dress well aren’t the people making dresses. that’s exactly what’s happening here
if our society doesn’t figure out a better way to distribute wealth, AI will help billionaires become trillionaires while average people become more exhausted than ever
Alex Gardner retweeted
MiniMax-H3-Image just dropped in early access!
I tested it by creating a floating city, then using that image as a reference for a market inside the same world.
One image gives you a location. A second can start telling its story.
The useful part was seeing which design choices carried across:
- Terracotta roofs and turquoise woodwork
- Giant fabric kites above the buildings
- Yellow railings connecting the spaces
The architecture changes, but the scenes share a visual language.
If you’re exploring a game or story concept, try this workflow:
Start with a wide view. Use it as a reference. Then move closer and show what people actually do there.
It gives you a practical way to explore a setting beyond its first impressive shot.
@MiniMax_AI #H3Image #ad
Alex Gardner retweeted
The lack of available energy will create a dog-eat-dog world in the AI industry. This is definitely not good for business.
当全世界的聪明人都在为近 4 万亿美元市值灰飞烟灭捶胸顿足时，硅谷风投大佬 Chamath 却掏出放大镜，找到了一个真正决定 AI 巨头生死的命门。
这位硅谷老狐狸抛出了一个暴论：眼下整场 AI 博弈的胜负，全看一单买卖。他指的，正是马斯克的 SpaceX 和 Anthropic 签下的那份算力合同，里面藏着一个“双向 90 天随时解约”的条款。
这事还得从这周的政治海啸说起。美国中期选举共和党遭遇了水门事件以来最惨的溃败，连红了几十年的德州都翻蓝了。州长们听见“数据中心”四个字就像见了瘟神，因为选民们早就受够了：这些庞大的机房除了替科技巨头赚钱，留给当地人的只有昂贵的水电费和噪音。
各地要是真给新数据中心按下暂停键，算力供给立马就得卡脖子。
就在这个节骨眼，硅谷的阶级成分瞬间清晰了。手里有现成地盘和电力的微软、谷歌、Meta、亚马逊，外加马斯克的 SpaceX，成了紧俏的“房东”；而像 Anthropic 和 OpenAI 这种得急需算力来迭代大模型的实验室，说白了，只是群没有房产证的“租客”。
在 Chamath 看来，这直接把马斯克送上了牌桌的主位。
手握着庞大算力，一旦数据中心审批放缓导致算力紧缺，现货价格可能会翻倍甚至涨到三倍。如果现货暴涨让原有长约变成了亏本买卖，SpaceX 随时可能直接动用“提前 90 天解约”的条款撕毁老合同，把算力按现货价重新计价。
到那时候，急需算力续命的 Anthropic 除了当场滑跪、认下翻倍甚至三倍的高价，根本没有第二条路可走。
就在这两天的富达路演上，面对机构把 IPO 估值从 2 万亿美元狠砍到 1.2 万亿的“骨折价”，CEO 达里奥为了保住上市，硬是诚恳地向买方投资者连说了三次“谢谢”。现在就开始赔笑脸了，真等到马斯克拔电源那天，他们更是连谈判的筹码都没有。
更损的是，这一刀要是真捅下去，等于当场给其他“房东”上了一堂生动的现实主义扫盲课，启发它们重新盘算手里的算力到底是留给自己用，还是拿去放高价租。而那些“二房东”们，为了不被踢出局，只能咬碎后槽牙，不计代价地去抢购市面上残存的算力。
听到这儿，旁边的 Jason 赶紧踩了脚刹车。他反驳说，马斯克跟 OpenAI 的奥特曼确实仇深似海，但跟 Anthropic 团队的关系好得像蜜月期，而且 Anthropic 还是他未来搞太空数据中心的好搭档。就算现货涨上天，马斯克也未必会干出直接掀桌子按现货价宰客的事，大家坐下来谈个“中间价”才更像人话。
但 Chamath 显然不相信硅谷的塑料兄弟情：“这跟关系好不好无关，跟谁是好公司坏公司也无关。这就是一张算账的表格，表格会直接给出答案。”
在这场近 4 万亿美元灰飞烟灭的魔幻风暴里，所谓的科技宏大叙事，剥开来看依然是最朴素的包租公哲学：只要电表转得够快、现货涨得够狠，再美好的革命友谊，可能也熬不过一张 Excel 表格，和一纸短短 90 天的逐客令。
Alex Gardner retweeted
SenseNova 6.8 Code2Video by @SenseTime_AI turns prompts into videos by generating editable code you can inspect and modify.
You describe a scene in plain language. The model translates that into editable code that defines visuals, motion, and timing.
A rendering system then runs that code to produce the video.
Since the source remains available, you can review or edit it and ask the model to update it in follow-up prompts.
The release includes weights, Skills, demo prompts, source code, and full creation traces for developers.
Explore the model on Hugging Face:
huggingface.co/sensenova/Sen…
Check out the project and star it on GitHub:
github.com/OpenSenseNova/Sen…
🤝 Paid partnership
Alex Gardner retweeted
Open-source models now handle about one in five tokens but account for only about one in 25 dollars of model spending, Menlo Ventures estimates.
From my February piece on China's token exports: "Chinese models dominate developer tooling on platforms like OpenRouter but have near-zero penetration in enterprise AI procurement."
Menlo's numbers cover all open-source models, and a US sample. They still show the same split between use and revenue. In that line I pointed to data sovereignty as the cause. Menlo puts its gap down to the difference in price between closed and open-source models. I think I gave price too little weight: the tokens get used, and each one bills for very little.
If a model wins the tokens but not the dollars, who keeps the value: the lab, or the cloud that serves it?
Alex Gardner retweeted
up to 84 t/s on a single rtx pro 6k
this is really interesting as I was not expecting a 65 billion parameter model to be performing as well as models over 10x its size on agentic coding tasks. but the intuition comes in when we think about how much of the parameters are just stored knowledge and memory, as opposed to this cognitive core that Karpathy described a little while back. in this case, it has less ability to store knowledge, and we can think of it as 65 billion active parameters for cognitive functioning. not to mention that this is a great recipe for super fast pre-training because they don't have to worry about the complexities of balancing and stabilizing an MoE.
this bad boy can fit on a single RTX Pro 6000 Blackwell at NVFP4 precision with the DFlash drafter getting probably really good performance for local setups. I bet there will be some interesting post-trains done on this model which will be extremely good at certain types of agentic coding tasks given that they said they had a light amount of post-training
claude code snapshot attached
Alex Gardner retweeted
If you do a simple trend extrapolation of this graph they should be at ~68% AL4 now. I would be really curious to hear if that is true. Seems important for @AnthropicAI to update this and for @OpenAI to also share more data!
Alex Gardner retweeted
AMD Local AI users, here's a vLLM update worth watching for your next build.
Octave KV, merged into vLLM, compresses the attention cache used while a model processes and generates text.
🎯 The author reports 4.6× the cached-token capacity with Qwen3.8-27B on 4× AMD RX 7900 XTX, compared with an FP16 cache.
That could allow longer conversations or more simultaneous requests within the model’s context limit.
This compresses the KV cache, not the model weights.
The author also reports comparable results on selected math and long-context retrieval tests.
🔗 Link in ALT.
ALT https://github.com/vllm-project/vllm/pull/59774
Alex Gardner retweeted
This is absolutely wild! Researchers at MIT CSAIL have demonstrated something that could fundamentally change how we build and optimize neural networks: "Weight-Space Programming".
Using GPT-6 Astra, they showed that an LLM can directly read, understand, and rewrite the weights of other neural networks, without gradient descent or retraining!
In one experiment, Astra discovered a reasoning-loop problem inside Qwen3.5-0.8B and fixed it by modifying just 0.016% of its 800 million weights. The result? Its MATH-500 score skyrocketed from 7.9% to 48.1%, while instruction-following performance nearly tripled!
Even more remarkably, Astra was able to construct neural networks from scratch, directly writing their weights to play video games. These networks outperformed models trained with 100 million steps of reinforcement learning!
Think about the implications. Instead of spending enormous amounts of compute retraining models, future AIs could potentially inspect, debug, and directly reprogram the neural networks of other AIs. We're only seeing the first experiments, but this is a fascinating new direction for AI research.
Alex Gardner retweeted
Will be common moving forward
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
Alex Gardner retweeted
Tencent Cloud vice-president Wu Qisheng explained the sales idea: an initial sale of WorkBuddy, Tencent's AI office agent, might bring only tens of thousands of renminbi, with more cloud business possibly following.
Tencent Cloud opened WorkBuddy to channel sales in May. The real question for this distribution effort is what those customers purchase later.
Tencent reports around 2,000 partners for WorkBuddy, its AI office agent, with more than 60% described as new.
A Shenzhen water-sector design consultancy’s procurement notice named Tencent Cloud as a candidate supplier for one year of WorkBuddy flagship accounts, at Rmb 23,760.
The notice gives no information on a signed contract, payment or the buyer’s previous purchases from Tencent.
I see this bid as an example of Tencent’s proposed entry into enterprise accounts. The larger cloud sale is the question.
My latest FlashPoint: hellochinatech.com/p/tencent…
Alex Gardner retweeted
This is true and I have always thought it was a big shame. Part of the reason (as Andrey notes in the replies) is that historically ABMs had not been taken to the data. But this is changing. Here is my colleague @RMaria_drc’s recent work bridging this gap:
arxiv.org/abs/2412.16591
ABMs x AI hold a lot of promise for economics, and it’s an active area of research for our team.
Alex Gardner retweeted
Ask orgs develop competencies around defining benchmark tasks, the more challenging problem will become developing *robust* tasks and verifiers. Tools should help eval engineers do this.
v0.25.0: new Harbor features to mitigate reward hacking
docs.harborframework.com/new…
Alex Gardner retweeted
A tool bug convinced several AI agents that breaking blocks in Minecraft was impossible.
YC Labs ran nine Bed Wars games: 10 Claude Opus 5.5 agents against 10 GPT-6.1 Sol agents. GPT won 6–3.
During the experiment, automatic tool selection failed when an enchanted sword was equipped.
A GPT agent found the workaround: manually equip an iron pickaxe, then call break_block. It verified that mining worked and messaged a teammate with the fix.
Several Claude agents concluded that block-breaking was impossible. One reasoned that buying shears was pointless because breaking was bugged, and kept guarding the base.
YC also observed Claude agents recording lessons in private memory files and repeating the mistakes. GPT agents broadcast warnings through DMs.
My takeaway for agent builders:
Test another tool state before declaring a capability unavailable. Verify the workaround. Share the finding with the other agents.
Your swarm needs to distinguish “this tool call failed” from “this task cannot be done.”
A workaround discovered by one agent only helps the team if the team learns about it.
labs.ycombinator.com/project…
Alex Gardner retweeted
.@TedEntertain's Notices are online at:
Meta - courtlistener.com/docket/720…
OpenAI -
courtlistener.com/docket/731…