Quick and dirty but an idea of how this works - back end configures a connection + front end for users.
Users ask questions in natural language and get a simple answer back. Works with mySQL/PostgreSQL.
Done in <2 hours with @langchain & @OpenAI
SQL is going to die at the hands of an AI. I’m serious.
@mayowaoshin is already doing this. Takes your company’s data and ingests it into ChatGPT. Then, you can create a chatbot for the data and just ask it questions using natural language.
This video demoes the output.
🤯
"Agents couldn't be trusted not to use MCP even if prompted not to, only removing this option fixed this"
"For repeatable tasks, a small harness is better than a larger general purpose"
If your harness doesn't follow your requirements, your model choice is irrelevant.
This week we read research from a team of academics that ran a software task across 7 agents and 5 models.
They found that in domains with a mature CLI ecosystem, agents without MCP baked in completed the task just as reliably and were 5-28x cheaper.
Full arXiv paper below
Ox Alpha has a 97% cache hit rate when I use it with @pidotdev
It's definitely a GLM family model. I've only seen one inference provider have a 90%+ cache hit rate and that's https://t.co/OjZcauYX4F
For reference Anthropic avg is 64% and OpenAI is 83% (upper bounds)
Possibly GLM. Basic identify yourself prompt got me this but I could not get this to replicate. It refused 100% of the subsequent identification requests 🤔
Ox Alpha is frontier. wtf. Seriously wtf. Tencent? Xiaomi? Really? It's more well-done than any other Chinese model. I swear, it's better than Kimi K3. It's good *in ways that Chinese models are typically horrible at*. It's FAST
I can half-believe it's a stolen Claude checkpoint.
Chinese RAM is starting to catch up with the memory used in Western gaming PCs.
Colorful has managed to run Chinese-made DDR5 at 9,000 MT/s, which is a very high speed for modern RAM.
It can also achieve good latency, chinese memory has been tested at DDR5-6000 with CL28/CL30, similar to the low-latency RAM commonly used in gaming PCs.
China’s biggest memory maker, CXMT, is also increasing production, and Chinese-made RAM is already appearing in some laptops.
The worst enterprise pitch you can give/hear is: we'll save you money, just sign here.
If you promise a company $1M in savings over X years, you can't also charge them $1M.
Ramp/Stripe have figured out is it's easier to sell $1 million in tokens than a $1M enterprise agreement.
We bought https://t.co/4WFUJAf1Cc.
Now we're building the best one.
Our customers buy quadrillions of tokens every month through Ramp, and we've run our own AI on this router for 3 years to keep costs down.
Today it's yours at https://t.co/4WFUJAf1Cc
Binface got 26.9% of total vote but his share was equal to 42.5% of Farage’s vote.
Farage: 63.3% → 11.7 points below poll
Binface: 26.9% → 6.9 points above poll
Protest candidate did better & fav did worse, that's a significant vibe shift.
Polling day for Clacton.
Reform - 70-73%.
Count Binface - 20%
Rest - 7-10%.
Binface wasn't going to get more than 50% but Clacton's demographic would have told you that for free months ago.
Reform W = back to square one, nothing changes politically -> investigation re-opened.
New level of underclass: who/when you get your intelligence from will affect your tariff 🙄
100% of what we have now is enough for 90% of tasks.
Compute bottlenecked for 2-3yrs
Who do you think can afford to pay more during peak hours? Hint: it's not gonna be their B2C users.
Not even an hour later, DeepSeek who proved frontier-quality AI can be built at a fraction of the cost are hiking their prices between 50%-1100%.
Still room for China to undercut here.
VC's being 61% of funding is crazy but it'll be the same playbook as every start up.
Your $20/m AI plan is subsided too, they will go bust or increase prices to stay in the game.
Enterprise adoption still early. 3-5 years avg for "transformation programs" to complete.
GL.
Polling day for Clacton.
Reform - 70-73%.
Count Binface - 20%
Rest - 7-10%.
Binface wasn't going to get more than 50% but Clacton's demographic would have told you that for free months ago.
Reform W = back to square one, nothing changes politically -> investigation re-opened.
Initially I predicted % share of all parties, this was before they stood down & let Count Binface contest the seat
Thesis remains: if major parties boycott, Farage still likely wins but Count Binface/protest candidate becomes the main danger.
Farage 46%
Count Binface 28%
The big players in AI desperately need enterprise adoption to be a success
It's the only way to recover billions in capex and provide a path to profitability
The big players in AI desperately need enterprise adoption to be a success
It's the only way to recover billions in capex and provide a path to profitability
VC's being 61% of funding is crazy but it'll be the same playbook as every start up.
Your $20/m AI plan is subsided too, they will go bust or increase prices to stay in the game.
Enterprise adoption still early. 3-5 years avg for "transformation programs" to complete.
GL.
If AI isn’t a bubble, you’ll lose your job. If AI is a bubble, you’ll lose your savings. Most likely, AI is currently in a financial bubble but technologically still on an exponential improvement curve, so you’ll first lose your savings and then lose your job. Good luck.
Everyone in corporate is a mental health activist until it actually matters.
That's why tribunals exist, not to determine liability but to determine reasonableness.
Harsh truth: being subject to a flawed disciplinary process can happen to anyone, it's you vs HR at that point.
Cursor == highly trusted brand by devs, trust earned over 3 years
Grok / xAI == uploaded your .env files to their cloud in secret, then pretended it was your fault for not setting a secret switch to avoid it
Weird to throw away the Cursor brand but go for it
"The moment Cursor feels like a captive xAI/Grok product rather than the best neutral coding environment, the acquisition starts destroying the asset it bought."
The draft policy incentivises venues to remove vertical drinking.
"removing vertical drinking opportunities counts in applicants' favour (D171)"
At best this is a contradiction of your own policy at worst it is deliberately misleading the public.
I love pubs. There is no ban on new licences if pubs don’t ‘ban’ standing. This is objectively wrong. It’s not what the policy under consultation says. Any policy to that effect would be illegal in any event. I’m not sure I can communicate this any more clearly.