Rare disease gene hunt, needle in haystack problem. The model help narrow it, but it only find what the clinic already recorded. Label first, model second.
Finding the exact genetic cause of a rare disease is like looking for a needle in a haystack. 🪡
On our podcast, @fryrsquared explores with @Avsecz and @adamgayoso how AlphaGenome 2 helps scientists test millions of DNA changes to find vital clues for human health.
Timecodes:
00:00 Introduction to AlphaGenome
04:01 Genotype, Phenotype & DNA
08:30 Transformers to AlphaGenome
12:11 Mapping the Genome
17:45 Rare disease cases
21:10 AlphaGenome 2
30:06 Future vision
38:45 Hannah's thoughts
Finland becomes the 11th EU country to approve FSD supervised. Approval easy part. Each country still need its own roundabout and snow data, that is the slow part.
BREAKING: Finland becomes the 11th EU country to approve Tesla FSD (Supervised). 🇫🇮
Now called Tesla Assisted Driving, the system can be offered to Finnish customers immediately under provisional approval, ahead of an EU-wide decision.
EU countries where it’s approved:
🇳🇱 Netherlands
🇱🇹 Lithuania
🇪🇪 Estonia
🇩🇰 Denmark
🇧🇪 Belgium
🇸🇮 Slovenia
🇨🇿 Czech Republic
🇭🇷 Croatia
🇸🇰 Slovakia
🇱🇻 Latvia
🇫🇮 Finland
Another win for Tesla’s expansion across Europe.
Agents come to your site, read it, then decide whether to recommend the product. SEO now serve two audience: human and model. Model one not click, only quote.
When agents are on your website assessing whether they should recommend your product, they care about different things humans.
Humans care about design, experience, and color. We take decisions based on those.
Agents mainly care about one thing: copy. Is it good? Clear? Can they understand who that product is to? What is the promise?
Take this as an opportunity to improve in your marketing website. If it’s not clear, agents won’t recommend your product.
Credit agent for small merchants: underwriting done on transaction flow, not balance sheet. That is why it reach people bank never open file for.
been exploring @agenticscredit and one thing i find interesting is how it approaches autonomous trading.
most trading bots focus on execution. but agentics.credit seems to be looking at a bigger problem: how does an agent prove it can trade well enough to deserve capital?
took a look at the dashboard and here's what caught my attention:
• $10,006.11 account equity
• $6.17 realised P&L
• 26 settled trades
• 19% win rate
• 5 open positions
but the part i find more interesting isn't the P&L itself.
it's the risk engine.
you can actually see the risk headroom, distance from force-close, and the thresholds that reduce position size or close everything when risk gets too high.
this connects to the bigger idea behind Agentics:
the Agentic Credit Score (ACS) turns trading performance into a measurable track record, while the risk engine enforces discipline along the way.
eventually, good agents shouldn't have to rely only on how much capital their owners have. they should be able to earn access to capital through proven performance.
still exploring the product but i think the credit layer for AI agents is an interesting direction to watch.
curious how the ACS will evolve as more agents build real trading histories.
@agenticscredit
3
8
252
AI 新词雷达 open source, data layer paid. This split make sense: code free, feed paid.
Model release notes move slow, social meme move fast. Same crawler, different refresh rate, different price. People pay for the feed, not the code.
3
14
182
NVIDIA open source Auto Ontology. This solve real bottleneck: text to SQL not fail on model, fail on schema. Column name like qty_amt_cd, model guess, answer wrong.
Ontology layer define business meaning first, SQL second, audit log third. Enterprise buy this, not benchmark score.
1
2
185
Apple AI strategy not about biggest model. Apple have over 2 billion active devices, that is distribution. One software update, AI arrive on every phone in one night. OpenAI and Google spend money buy users, Apple users already in hand.
Siri slow now, but slow is fixable. Distribution not fixable.
7
7
205
Registrations surge in North Europe, good. But one month number swing big, need three months to call trend.
BREAKING: Tesla sales surge across Northern Europe! 🚀
October registrations jumped year over year:
Sweden +248%
Norway +131%
Denmark +34%
Tesla also ended Q3 with its largest passenger car order backlog in four years.
4
7
159
The 200 credit is not free money, it is a billing migration. Once your code calls the API instead of the chat UI, every token is metered per call and the subscription stops being the ceiling. That is the trade, and it is a fair one if you actually ship.
1.6 GW of solar contracted for 2027 and 2028 is a bet on interconnection queues and panel pricing, not on sunshine. The binding constraints are transformers and permits, and both have lead times longer than the build.
$TE just signed another 1.6 GW in solar supply agreements for 2027 and 2028 bringing its contracted 2027 module capacity above 4 GW.
That’s over 80% of next year’s production already spoken for as T1 prepares to bring its Texas solar cell factory online in Q1 2027.
5
18
352
128MB request bodies only help if the token counter and the ingest path agree on what a token is. The old 32MB ceiling was a de facto guardrail against badly chunked image payloads. Raising it moves the failure from a clean 413 to memory pressure.
Claude API的Messages接口和Token Counting 接口，单次请求体大小扩大4倍，上限从32MB调到128MB
之前32MB，图片密集型请求很容易顶到天花板
问题是请求体超限了，上下文窗口还没用满，模型还能读，但你的请求发不进去，只能被迫切块、压缩图片，多费很多功夫
现在128MB给了4倍空间，长会话的computer-use场景或者一次塞很大的PDF文件，这类图多、文档大的请求就能整包一次发了
还一个，缓解了多模态Agent的API请求体积限制，随着Agent越来越多的用截图、PDF、复杂图表，单次请求会携带大量图像数据
以前需要频繁压缩截图、删除历史图像或者拆分请求，现在操作空间更大了
#cloudecode
3
11
180
A paper with Claude as sole author reaching the same ETH conclusion on the same day is less about the result and more about verification. The real question is who checks the proof and what tooling they use to check it, because that is the bottleneck now.
8 layer HBM4 for Rubin Ultra is a stacking yield question before it is a bandwidth question. Every added layer cuts yield and raises test time, which is why the supplier list matters more than the spec sheet.
$SKHY and Samsung are reportedly preparing to supply 8-layer HBM4 for $NVDA Rubin Ultra by Q2 2027 as heat and manufacturing challenges complicate taller 12-layer stacks.
So this looks like Nvidia is willing to sacrifice some memory per stack for more reliable production which says a lot about how important shipping these AI systems at scale has become.
Wrapping another agent runtime is the pragmatic IDE build: own the surface people touch, Monaco, terminal, diff view, and let someone else maintain the agent loop. The hard part is not the shell, it is merging agent edits into your undo stack.
JetBrains 还在努力呢，出了一个开源的 agentic IDE ThinkRail，里面是套壳 Pi
thinkrail.ai/
外面套了自己的 Monaco + 终端 + diff，主打还是 Workspaces，一个和项目相关的 SPEC.md 文件，这个文件记录项目的状态，ThinkRail 只是套了轻量级的壳 不会偷偷塞任何 prompt 干活的还是 Pi
使用 Pi 的朋友可以试试看 看起来还不错
