Jim Fisk retweeted
This is NH School Facts 2.0.
A completely rebuilt database, with every number from government sources, and every source linked.
Look up your town, district, or SAU for enrollment, spending, property taxes, test scores, and more.
nhschoolfacts.com
A completely rebuilt database, with every number from government sources, and every source linked.
Look up your town, district, or SAU for enrollment, spending, property taxes, test scores, and more.
nhschoolfacts.com
6
17
69
1,377
Jim Fisk retweeted
Build Your Own #Robot Dog With This ESP32-Powered Board
by @Alacritic_Super
#Robotics #Engineering #Innovation #EmergingTech
by @Alacritic_Super
#Robotics #Engineering #Innovation #EmergingTech
14
47
535
21,164
My takeaway: self-host your fonts with the app itself. This probably holds for all CDNs
Nobody warns vibe coders about this part.
You can ship an app in a weekend.
You can also break 6 laws in that same weekend without knowing it.
Here's what to check before you launch:
1.) Kids
If signup doesn't ask for age and a child under 13 joins, COPPA fines can reach $53K per child.
2.) Fonts
Loading Google Fonts from Google's servers sends visitor IPs to Google. A Munich court ordered a site to pay one visitor €100 for this under GDPR.
3.) Session replay
Recording what users type without consent can count as wiretapping under California's CIPA, at up to $5K per violation.
4.) Launch emails
No unsubscribe link or postal address breaks CAN-SPAM, at up to $53K per email.
5.) Subscriptions
If renewal terms aren't right next to the button, California can treat renewals as unconditional gifts you have to refund.
6.) User uploads
Without a registered DMCA agent, which costs $6, you lose safe harbor. Statutory damages can reach $150K per willfully infringed work.
None of these cost much to fix.
All of them cost a lot to ignore.
Paste this into Claude before launch:
"Check my app for: missing age gate, externally loaded fonts, session replay without consent, marketing emails without unsubscribe and address, renewal terms not beside the subscribe button and no DMCA agent. Fix what you can and list what I need to do myself."
Not legal advice. Talk to a lawyer for your situation.
0
0
3
78
Jim Fisk retweeted
DeepSeek doesnt need a subscription. The weights are MIT. Every other lab should be embarrassed.
You pay per token on their API, or you download V4.1 Flash and never pay per token again. I run it on my 4x 3090 at 57 tok/s and nobody can cap me.
Now look at the rest of the market this month. SemiAnalysis caught Anthropic A/B testing usage limits, and some accounts got about 20% less monthly value without a word. My Kimi 5 hour limit was gone in 10 minutes last week. Moonshot at least apologized and reset everyone. Anthropic didnt even tell people.
With a subscription the company decides how much you get, and it can change that any day it wants. With MIT weights you decide. Forever.
Thank you @deepseek_ai. Everyone else, take notes.
x.com/Da7_Tech/statu…
You pay per token on their API, or you download V4.1 Flash and never pay per token again. I run it on my 4x 3090 at 57 tok/s and nobody can cap me.
Now look at the rest of the market this month. SemiAnalysis caught Anthropic A/B testing usage limits, and some accounts got about 20% less monthly value without a word. My Kimi 5 hour limit was gone in 10 minutes last week. Moonshot at least apologized and reset everyone. Anthropic didnt even tell people.
With a subscription the company decides how much you get, and it can change that any day it wants. With MIT weights you decide. Forever.
Thank you @deepseek_ai. Everyone else, take notes.
x.com/Da7_Tech/statu…
I'm slowly starting to get why DeepSeek doesn't offer a subscription.
They actually care about transparency and giving you a solid experience without that feeling of getting scammed like you do with Kimi or Zai.
29
32
683
22,722
Jim Fisk retweeted
The next great "language for agents" should be:
- As fast as Rust
- As little design space as Go
- Have the strictest type/verification system ever written
Bend seems like an interesting first effort
- As fast as Rust
- As little design space as Go
- Have the strictest type/verification system ever written
Bend seems like an interesting first effort
510
220
6,328
724,790
Jim Fisk retweeted
If you’re asking why Anthropic would do this and why it’s not normal, let me explain.
Anthropic did this at the request of nervous investors who wanted to hide major OpEx that show Anthropic has basically no chances of profitability in the future.
In a normal environment, no early investor would EVER want this, because they would be substantially diluted. In this case, the IPO target is priced so hilariously stupid that it’s a far better tradeoff to the early investors to commit PERFECTLY LEGAL “fraud” to hide the fact Anthropic doesn’t have any path to profitability. $108 billion dilution at IPO is only 5% of equity lost. The investors can easily cover that with puts at IPO to insure their gains.
It only happens in major bubbles. Brought to you by $NVDA
Anthropic did this at the request of nervous investors who wanted to hide major OpEx that show Anthropic has basically no chances of profitability in the future.
In a normal environment, no early investor would EVER want this, because they would be substantially diluted. In this case, the IPO target is priced so hilariously stupid that it’s a far better tradeoff to the early investors to commit PERFECTLY LEGAL “fraud” to hide the fact Anthropic doesn’t have any path to profitability. $108 billion dilution at IPO is only 5% of equity lost. The investors can easily cover that with puts at IPO to insure their gains.
It only happens in major bubbles. Brought to you by $NVDA
37
179
1,311
111,862
Jim Fisk retweeted
Prosperity and peace are imminently attainable for everyone in this country. We need merely to transcend the parties that keep us fighting each other, throw off the corporate and foreign influence on our government, and return to the principles laid out in our Constitution.
1,422
3,957
21,231
330,819
Jim Fisk retweeted
@Yuchenj_UW I think we’re going to see a lot of stranded compute in partially complete data centers
1,039
812
15,087
5,980,858
Jim Fisk retweeted
READ: if everyone used DeepSeek’s ultra efficient v4.1 flash model there would be such a gargantuan compute glut already that we could just start canceling a bunch of the future compute commitments.
The reality: this isn’t going to happen overnight but this is the direction we are going and we haven’t even stopped the logarithmic efficiency improvements.
Yes token consumption will grow exponentially but there are so many supply exponentials that will completely overwhelm demand.
COMPUTE GLUT.
The reality: this isn’t going to happen overnight but this is the direction we are going and we haven’t even stopped the logarithmic efficiency improvements.
Yes token consumption will grow exponentially but there are so many supply exponentials that will completely overwhelm demand.
COMPUTE GLUT.
Western team have independently verified that DeepSeek's performance numbers of 38$ dollar revenue per GPU at DeepSeek's rocket bottom prices is true.
This is a momemnt when you update your priors. At this rate with just 1 GW, you can make a revenue of 150B per year.
Even with e 100B per year. If you use half the compute to come online by 2027 end, you will have 25GW for inference with potentially 2.5T revenue at DeepSeek rates and 10T at USA lab rates (world GDP is 125 T, 10T is 8% of it)
You answer for your-self are we overbuilding!!!
11
11
113
10,662
Jim Fisk retweeted
I just built an ESP32 classroom with Astra
no need to buy a starter kit just to figure out what the pins do, start in your browser:
> explore a 3D ESP32 and hover to learn what each component does
> follow highlighted pins and step-by-step wiring guides
> experiment with LEDs, senpiano
> go from interactive demos to wiring diagrams and Arduino code
16 lessons. Free & open source. Desktop & mobile.
Try it in reply↓
no need to buy a starter kit just to figure out what the pins do, start in your browser:
> explore a 3D ESP32 and hover to learn what each component does
> follow highlighted pins and step-by-step wiring guides
> experiment with LEDs, senpiano
> go from interactive demos to wiring diagrams and Arduino code
16 lessons. Free & open source. Desktop & mobile.
Try it in reply↓
7
26
241
13,881
Jim Fisk retweeted
If this PHP vulnerability allows remote code execution, can’t we just use it to patch every vulnerable server?
It’s not an exploit. It’s an unsolicited update service.
It’s not an exploit. It’s an unsolicited update service.
0
1
12
497
I was as skeptical as it came regarding coding agents, and imo rightfully so early on (it used to produce garbage). It's entirely different now, and the only way you wouldn't be able to see that is if you're actively avoiding trying it in earnest
the coding agent haters are getting dunning-kruger'd so hard right now
do you really think that your engineering ability is better than the models at this point?
it is very very likely that fable is better than you in every branch of software
it has been trained on all the code of the internet, it knows almost every trick in the book, all the design patterns of every domain, every single low-level detail, every bit-twiddling hack, it's read every single blog post from every cracked engineer that ever existed
do you really think you can beat it? do you really think you can be more creative than it?
unless you are 99th percentile expert in some niche domain of software, it simply is not possible. sorry.
0
0
0
39
You had me at "Runs on plain SQLite"
China has killed the entire vector database industry.
They open-sourced TencentDB Agent Memory. It gives any AI agent long-term memory that runs 100% locally.
No Pinecone. No cloud APIs. No repeating yourself every session.
- 61% fewer tokens
- PersonaMem accuracy: 48% → 76%
endencies
- Runs on plain SQLite
Most memory systems compress your history into an opaque vector pile. when recall goes wrong, you're guessing. this one doesn't compress, it builds a semantic pyramid.
L0 Conversation → L1 Atom → L2 Scenario → L3 Persona.
Short-term state gets encoded as a Mermaid graph in your agent's context. verbose tool logs get offloaded to disk. when the agent needs proof, it drills back via node_id to the exact raw log.
no lossy compression. every layer is readable markdown you can just open and inspect.
5.1k stars. 100% Open Source.
0
0
0
42
Jim Fisk retweeted
The Linux Kernel is removing the requirement of tagging AI generated code submissions.
Over the last 10 months, we’ve been able to track the growing rate of AI coding in the kernel (with over 20% of all code submissions being AI generated, so far in October)… we know this required “Assisted-by” AI tag.
With the removal of this requirement, we will no longer be able to effectively track the level of AI development within Linux.
This change was agreed upon (by the majority of kernel devs) at the Linux Maintainers Summit, held last week in Prague.
The stated reasons for removing the “AI declaration” requirement are varied, but come down to these three arguments:
1) Usage of AI for coding the Linux kernel is becoming so widespread, that declaring it is redundant. (One developer likened it to declaring which text editor you used.)
2) Linux Kernel developers feel that most developers who use AI… are not actually following the rule of labeling their submissions as such anyway.
3) Other developers feel like it is “free advertising” for various AI platforms.
This change has not been officially merged yet, but (considering it was already agreed upon, and seems to have the support of top level maintainers at the Linux Foundation) that is expected to happen shortly.
Over the last 10 months, we’ve been able to track the growing rate of AI coding in the kernel (with over 20% of all code submissions being AI generated, so far in October)… we know this required “Assisted-by” AI tag.
With the removal of this requirement, we will no longer be able to effectively track the level of AI development within Linux.
This change was agreed upon (by the majority of kernel devs) at the Linux Maintainers Summit, held last week in Prague.
The stated reasons for removing the “AI declaration” requirement are varied, but come down to these three arguments:
1) Usage of AI for coding the Linux kernel is becoming so widespread, that declaring it is redundant. (One developer likened it to declaring which text editor you used.)
2) Linux Kernel developers feel that most developers who use AI… are not actually following the rule of labeling their submissions as such anyway.
3) Other developers feel like it is “free advertising” for various AI platforms.
This change has not been officially merged yet, but (considering it was already agreed upon, and seems to have the support of top level maintainers at the Linux Foundation) that is expected to happen shortly.
36
78
731
50,333
Jim Fisk retweeted
Local AI is evolving REALLY fast
Qwen 3.8 27B was released just 2 months ago and was smarter than Opus 4.6
But you needed powerful/expensive hardware to run it
MiniCPM-V 4.7 is as powerful and:
- only 3B active parameters (!)
- it runs even on an iphone
- better than models meters
So you can run a model on a phone that is better than the most powerful one available 6 months ago.
The future is local.
Qwen 3.8 27B was released just 2 months ago and was smarter than Opus 4.6
But you needed powerful/expensive hardware to run it
MiniCPM-V 4.7 is as powerful and:
- only 3B active parameters (!)
- it runs even on an iphone
- better than models meters
So you can run a model on a phone that is better than the most powerful one available 6 months ago.
The future is local.
🚀 Introducing MiniCPM-V 4.7, our new open-source MLLM series!
Advancing the Performance–Efficiency Pareto Frontier.
It brings strong multimodal capabilities to iOS and Android devices with extreme efficiency.
Two new models:
🔹 MiniCPM-V 4.7 35B-A3B — Bring SOTA performance to y introduced LES (Lookahead Expert Scheduling) technique, a new milestone of on-device multimodal AI.
🔹 MiniCPM-V 4.7 1B — Lightweight, efficient multimodal AI. Easier to deploy.
⚡ What's new?
SOTA performance: Competitive results across 34 benchmarks spanning multimodal understanding, STEM reasoning, OCR and long-video comprehension.
Fast on-device inference: LES predicts and preloads expert weights to reduce inference stalls, boosting 5× speedup on iPad Pro M4 (16GB) with byte-for-byte identical outputs. It enables MiniCPM-V 4.7 35B-A3B to achieve 17+ tokens/s on iPad Pro M4 (16GB) and 13+ tokens/s on iPhone 17 Pro Max, while costing only 3GB memory 🤯
Efficient visual processing: 4× and 16× hybrid visual token compression. Even at 16× compression, 35B-A3B scores 90.4 on OCRBench, 5.2 points ahead of Qwen3.6 35B-A3B.
Easier customization: We introduce Self-Replay Fine-Tuning, a simple way to customize new target capabilities at low cost while maintaining general performance.
Broad hardware support: Day-0 adaptations for Apple, Intel, AMD, Arm and Rockchip, with many more platforms supported through FlagOS!
Download, deploy, and build something we haven't imagined yet. 👀👇
🔗 GitHub: github.com/OpenBMB/MiniCP…
🤗 Hugging Face: huggingface.co/collections/op…
90
197
2,580
190,445
